Sustainability-integrated value stream mapping with process mining

ABSTRACT

Value stream mapping is a well-established tool for analyzing and optimizing value streams in production. In its conventional form, it requires a high level of manual effort and is often inefficient in volatile and high-variance environments. The idea of digitizing value stream mapping to increase efficiency has thus been put forward. A common means suggested for digitization is Process Mining, a field related to Data Science and Process Management. Furthermore, adding sustainability aspects to value stream mapping has also been subject to research. Regarding the ongoing climate crisis and companies’ endeavors to improve overall sustainability, integrating sustainability into value stream mapping must be deemed equally relevant. This research paper provides an overview of the state of the art of Process Mining-based and sustainability-integrated value stream mapping, proposes a framework for a combined approach, and presents technical details for the implementation of such an approach, including a validation from practice.

KEYWORDS:

• Process mining
• value stream mapping
• sustainability
• manufacturing

1. Introduction

This paragraph provides an overview of the topic of digitized sustainability-integrated value stream mapping, the structure of the paper, the reviewed research questions, and the research methodology.

1.1. Research topic and structure of the paper

Digitization is reshaping the manufacturing industry at all levels. From collaboration in production networks to optimizing specific technologies – digitization facilitates more effective and efficient production processes.

Optimization methods formerly requiring physical presence and relying on observations can now be implemented through software, allowing a more objective perspective and enabling continuous real-time analysis. The well-established value stream mapping methodology as a pen-and-paper-based methodology requires great effort to provide a holistic view of production processes for selected product families (Erlach, 2020). The manual efforts required for conventional value stream mapping render the methodology inefficient in highly volatile environments or companies with a diverse product portfolio. Theoretical concepts of digitizing the value stream mapping methodology have thus been put forward, mostly implementing Process Mining as a means of digitization, but have not been validated in practice using actual and diverse production data (Horsthofer-Rauch et al., 2022). Furthermore, integrating sustainability-related aspects into digitized value stream mapping approaches has gained little attention. Yet, sustainability-integrated value stream mapping approaches have been discussed for the last 20 years and are once more legitimized by the climate crisis.

The idea behind Process Mining-based sustainability-integrated value stream mapping is to enable continuous analyses of all product value streams in real-time and enhance the insights by considering additional data. By providing such insights, production planning and control activities can be efficiently and effectively supported. Compared to the conventional pen-and-paper-based approach, the digital analyses of all value streams can be executed continuously at acceptable costs (Knoll, Waldmann, Reinhart, 2019). Additionally, deploying digitally stored data facilitates a more holistic and objective view of production. The consideration of different sustainability aspects in the analyses supports the integration of these aspects in operations management through a well-established communication medium (Lorenzon dos Santos et al., 2019). It ensures the equal consideration of economic, ecological, and social aspects for optimization.

This paper provides an overview of a novel general framework for digitized sustainability-integrated value stream mapping using Process Mining and explains relevant technical details for the practical implementation. It aims to support universal practical applicability, which has so far been lacking in research and is essential, especially in today’s highly volatile, high-variance production environment. Furthermore, the combination of digitized and sustainability-integrated value stream mapping is presented to account for the need to integrate sustainability aspects into production. A single use case in a company with a diverse product portfolio is presented, and its implications are discussed. The research was conducted within the publicly-funded research project ProVSA.

The paper is structured as follows. The remainder of Chapter 1 presents the underlying research questions and research methodology employed to structure the research project and define effective goals for research and development. The second chapter provides a generic overview of value stream mapping and Process Mining in literature and practice to introduce the topics. Chapter 3 contains the state of the art of research in sustainability-focused value stream mapping, Process Mining-enabled value stream mapping, and the intersection of the topics. In Chapter 4, the approach for Process Mining-enabled sustainability-focused value stream mapping is introduced and detailed with regard to technical implementation. Chapter 5 provides information on the results achieved within the company case study and presents the defined framework and technical details for its implementation into practice. In Chapter 6, the results from the practical validation are discussed, limitations are identified, and an outlook on future research is provided.

1.2. Research questions

The research project was carried out by scientific as well as industrial partners. The overarching goal was to research and develop a solution for digitizing and enhancing the conventional value stream mapping methodology using Process Mining. The following research questions were approached:

1. What data is required to carry out sustainability-integrated value stream mapping, and how does this data need to be pre-processed?

2. Which types of analysis support digitized value stream mapping based on Process Mining, and how can these be implemented?

3. How can users be supported in gaining relevant insights through a digitized, sustainability-integrated value stream mapping approach?

1.3. Research methodology

The research was conducted following a pragmatic research philosophy, through which the stance was taken that concepts must support action and that research starts with a practical problem (Saunders et al., 2019). The research methodology was chosen based on the preliminary studies that preceded this project. The prestudies included an unstructured review for research clarification and interviews with practitioners, resulting in the choice of a pragmatic, action-focused research philosophy.

To best support action, members from research, development, and practice were involved, bringing together experts from information technology and production technology. Involved in the project were the Institute for Machine Tools and Production Engineering (iwb) of the Technical University of Munich (TUM), Celonis SE, a cloud provider and industrial leader in Process Mining applications (Kerremans et al., 2023), and HAWE Hydraulics SE, a manufacturer of hydraulic aggregates and components.

A Design Thinking approach was used to ensure the support for practice and to account for demands in information technology research and development projects. Design Thinking is an iterative methodology that produces user-oriented solutions to complex problems. Design Thinking allows the development of solutions for products, processes, systems, and objects. The methodology requires close collaboration between researchers, developers, and potential users (Uebernickel et al., 2020). The Design Thinking methodology was used to lead the project team in their research, development, and evaluation. To fully integrate the views of the user into the research and development process, so-called User Stories were developed based on the identified relevant Personas. Whilst Personas are a common method used in Design Thinking processes to supply descriptive models, i.e. an archetype, of the potential user (Uebernickel et al., 2020), User Stories are rather rooted in Agile practices. User Stories provide statements from the standpoint of potential users that translate into distinct descriptions of desired functionalities that are valuable to the user (Cohn, 2004). These User Stories ultimately create a common view of what is important within the subject of the study and serve as a means for discussion between researchers, developers, and practitioners.

To validate the methodology and technical concepts, a use case at HAWE Hydraulics SE was considered.

2. Background

In the following, an overview of value stream mapping and Process Mining is given, discussing the fundamentals, and current developments in science and practice. The chapter aims to provide a common understanding.

2.1. Value stream mapping

The methodology of value stream mapping was introduced by Rother and Shook in 1999 and is based on the Lean Management philosophy. The idea is to analyze the value stream of a certain product or product family to identify inefficiencies and opportunities for improvement within production. Visualization of production processes, material flow, and information flow is the foundation for further analysis. The visualization is captured in the so-called value stream map. In its conventional form, value stream mapping is a pen-and-paper-based methodology and requires a physical presence on the shop floor to obtain data (Rother & Shook, 2018). The manual nature of conventional value stream mapping creates high costs, especially in more complex environments (Erlach, 2020). Generally, value stream mapping starts with product family formation. To reduce complexity, a representative from the product family is chosen for the analysis. For this representative, the value stream analysis is carried out, and an optimal value stream is designed based on the insights on wasteful, i.e. inefficient and ineffective, behavior from the value stream analysis. Insights and proposals for action are then leveraged for all products of the product family, i.e. their respective value streams (Erlach, 2020). Figure 1 shows the relationship between the different steps of the value stream mapping methodology, as described above.

Figure 1. Steps of the value stream mapping methodology (Rother & Shook, 2018).

In recent years, approaches for digitizing value stream mapping have been put forward to account for the increased costs of manual value stream mapping due to rising process complexity and higher product variance (Urnauer et al., 2021). These approaches make use of readily available data. The utilization of automatically recorded data not only allows a more efficient but also more effective value stream mapping approach, as more and more accurate data can be considered. A common denominator for digitized value stream mapping approaches is the deployment of Process Mining as the means of digitization due to the process-centric nature of the data analytics method (Horsthofer-Rauch et al., 2022).

Independent of the digitization approach, another extension of value stream mapping has gained popularity in academia. Many approaches for sustainability-integrated value stream mapping have been presented, the most cited by Faulkner and Badurdeen (2014). Sustainability-integrated value stream mapping approaches take into account not only the common economic but also ecological and social aspects (Faulkner & Badurdeen, 2014). The approach of integrating sustainability aspects into a well-established optimization tool lowers barriers when introducing sustainability controlling in production (Lorenzon dos Santos et al., 2019). The benefits of digitizing the sustainability-integrated value stream mapping approach to lower required efforts for data collection and data processing were put forward by Scheder et al. (2023), Horsthofer-Rauch et al. (2021), and Phuong et al. (2018).

2.2. Process Mining

Process Mining aims to discover, monitor, and improve real processes by extracting knowledge from event logs (W. van der Aalst, 2016). Process Mining is used to visualize how data-driven processes in companies truly run. It is designed to draw a comparison to the intended process and identify inefficiencies. For this, Process Mining extends Machine Learning and Data Mining by the process perspective and forces the confrontation between reality, mapped in event data recorded from real events, and the process model. In the context of production, all three Process Mining types, Discovery, Enhancement, and Conformance Checking, can be deployed. To automatically create process models without a priori knowledge about the process, Process Mining Discovery techniques are used. Discovery allows the extraction of a process model, e.g. of production processes, and can be performed offline, i.e. after the fact, or online, i.e. live. The former aims at a better understanding of the process and the review of improvement opportunities. The latter enables support for ongoing operations in the sense of near-to-real-time process monitoring (W. van der Aalst, 2016). Enhancement facilitates the adaption of a process model and the extension with further performance metrics. Lastly, Conformance Checking allows the comparison of processes found in real event data against an existing process model (W. van der Aalst, 2016). The deployment of Process Mining in the context of production has been widely discussed in scientific publications, and application scenarios differ (Knoll, Waldmann, Reinhart, 2019; Urnauer et al., 2021; Ziegler et al., 2019).

A novel approach to Process Mining is object-centric process mining (OCPM). It assumes, going beyond the well-established case-centric approach of Process Mining, that processes are not happening in isolation. It describes that multiple objects interact with each other by sharing events. OCPM allows for a multidimensional and, thus, more holistic picture of processes within companies by overcoming the issues of divergence (events within the same case cannot be separated) and convergence (events related to different cases are duplicated) (W. M. van der Aalst et al., 2019).To make use of OCPM, a new kind of event log, called an object-centric event log (OCEL), was introduced. An object-centric event log enables event notations to have a relation to multiple objects (e.g. sales order items, production orders, etc.), whereas, in the traditional approach of Process Mining, only case-centric event data is analyzed (Ghahfarokhi et al., 2021).

Within the research project ProVSA, we compiled a detailed understanding of the dependencies between different objects in production processes, such as machines, labor, production orders, materials, or process plans. The approach selected in this work is considered a foundational step to OCPM, as the developed object schema models the relationships between the crucial components in manufacturing. In this work, the interrelationships between different materials in the production process are mapped directly by leveraging the bills of materials. Hereby, the full value stream between procuring raw materials and selling finished products is represented, which enables end-to-end analytics and insights. This facilitates the transition to OCPM in a later stage.

3. State of the art

To integrate and build a framework upon current scientific findings, a structured literature review was conducted following the approach by Vom Brocke et al. (2015). Literature reviews were conducted on the topic of Process Mining-based value stream mapping, sustainability-integrated value stream mapping, and the combination of these approaches. Employing the classification by Vom Brocke et al. (2015), all literature research was conducted following a sequential process, using bibliographic databases and publications and conducting a keyword search. The database used for identifying relevant publications via keyword search was SCOPUS. SCOPUS was used as it only lists peer-reviewed publications, enables structured searching and filtering, provides more results on the topics than other databases to which we had access, and extensively includes German content, which is especially relevant in the context of Process Mining-enabled value stream mapping. The search string was used to analyze article titles, abstracts, and keywords of publications. English and German publications were considered. No time frame was instated to ensure all relevant literature was reviewed independent of publication dates. Publications were considered if they were focused on discrete production and did not only present a single use case in industries with unique equipment, such as the textile or process industry.

The findings are presented in Chapters 3.1–3.3. The conclusions from the state of the art are discussed in Chapter 3.4.

3.1. Process Mining-based value stream mapping

To identify relevant scientific approaches for Process Mining-based value stream mapping, the search string (‘Process Mining’ AND ‘value stream m*’ AND NOT sustain*) was used, where the asterisk* allows different endings of a word. The search yielded 13 results, of which six seem relevant after scanning the title and abstract. Of those six possibly relevant publications, five remain relevant after reading the full text. The relevant publications are presented in descending order of their publication in the following.

Horsthofer-Rauch et al. (2022) reviewed digitized value stream mapping approaches and highlighted the advantages as well as the required development of Process Mining as a means of digitization for value stream mapping. According to the categorization of the authors, all three Process Mining types, Discovery, Enhancement, and Conformance, can be used to support different aspects of the value stream methodology (cf. Figure 2). Furthermore, relevant areas of development are presented in the publication, which are the definition of data requirements and a methodology for data preparation, the analysis of visualization requirements and capabilities, and the possibility of an extension of the value stream mapping methodology (Horsthofer-Rauch et al., 2022).

Figure 2. Matching of value stream mapping aspects and Process Mining types (Horsthofer-Rauch et al., 2022).

Rudnitckaia et al. (2022) proposed a combination of Process Mining and value stream mapping for bottleneck identification. Process Mining is used to create a descriptive process model to depict the value stream and as a starting point for bottleneck analysis. Regarded key performance indicators are lead time and waiting times. The shop floor control number of a product serves as the case identifier (ID), and the distinct operations constitute the activities. The start and end times of operations are used as timestamps. The authors suggest enriching the event log with further data from production to provide additional perspectives (Rudnitckaia et al., 2022).

Nawcki et al. (2021) presented a case study on the use of Process Mining to support value stream mapping in quality management. The authors constituted that Process Mining can help reduce the required efforts for analyzing large amounts of data and complex processes and that deviations between the perceived processes and the processes as executed in reality. Information on the retrieved data and the implementation for the use case are provided (Nawcki et al., 2021).

Urnauer and Metternich (2019) suggested combining Process Mining and value stream mapping to facilitate a more objective perspective. Process Mining is viewed as a complementary database for the value stream mapping methodology by making use of readily available manufacturing data. The generated process model and respective indicators are integrated into the regular value stream mapping methodology as a means of discussion and control of manually recorded information and assumptions. Details on required data and implementation are generic. The concept was not validated (Urnauer & Metternich, 2019).

Ziegler et al. (2019) described a dynamic value stream mapping approach using Process Mining. The authors proposed that Process Mining can support the analysis and visualization of implicitly available knowledge trapped in readily available transactional data. To enable Process Mining-based value stream mapping, data first needs to be identified in relevant IT systems and then extracted. Process Mining is then applied to provide near real-time value stream maps. Details on required data and implementation are generic. The concept was not validated (Ziegler et al., 2019).

Knoll, Waldmann, Reinhart (2019) identified the need for a digitized value stream mapping in internal logistics due to its complex and dynamic nature. The authors adopted the method of multidimensional Process Mining to use in internal logistics to ultimately identify waste in value streams. Transfer orders serve as case IDs and event logs contain information on storage location activities and timestamps. Discovery and Conformance techniques are employed to assess value streams and their adherence to the target process. Further, a performance analysis based on the insights from these techniques is conducted. Details on required data are provided in the approach, which was finally validated at an automotive company (Knoll, Waldmann, Reinhart, 2019).

Klenk (2019) also proposed an approach to automatically derive a value stream model from readily available data using Process Mining. To validate the Process Mining-based results or adjust deviations in the database or the physical production system, a conventional value stream mapping is conducted. Klenk suggested further including IT processes that influence production processes in the value stream map to provide a fully holistic view. Details on required data and implementation are generic. The concept was not validated (Klenk, 2019).

The topic has further received attention in practice. Various consulting companies offer Process Mining-based value stream mapping in their service portfolio yet do not provide public information on the subject. The provider of Process Mining solutions, MPM also offers process mining-based value stream mapping in its product portfolio. The solution seems to be a tailored service rather than a generic approach. The approach is based on the software QlikSense.

All presented publications with a focus on production either present concepts or specific use cases lacking detail for required data that is needed to implement the approach in other companies. The topic of sustainability is not touched upon in either of the full texts of the presented approaches. Nevertheless, the available concepts can be adapted to derive an efficient and effective approach for Process Mining-based value stream mapping.

3.2. Sustainable value stream mapping

To identify relevant scientific approaches for sustainable value stream mapping, the search string (sustainable AND (ecologic* OR environment* OR societ*) AND ‘value stream m*’) was used. The search yielded 133 results, of which 18 remained relevant after analyzing the title, abstract, and, finally, the full text. Application case studies and publications focused on specific industries were excluded. As the focus of this paper is to support the digitization of value stream mapping using Process Mining while integrating sustainability aspects and not to discuss the scientifically well-established methodology of sustainability value stream mapping itself, only a few representatives are presented. Thus, whilst the literature research was comprehensive, a representative coverage is chosen for this publication. With the representative coverage, we aim to present a shortlist of approaches that were relevant to our research due to their input on indicators and calculation formulas, methodology, and digitization. The relevant publications are presented in descending order of their citation according to SCOPUS.

The most cited publication on sustainability value stream mapping stems from Faulkner and Badurdeen (2014). The authors present a holistic methodology for sustainability value stream mapping by extending conventional value stream mapping with environmental and social metrics and providing insights into the practical implementation. Furthermore, an adapted visualization for mapping environmental, social, and economic metrics alike is presented. The methodology was validated in a use case (Faulkner & Badurdeen, 2014).

Jamil et al. (2020) reviewed existing approaches and proposed a systematic sustainability value stream mapping approach deploying the define, measure, analyze, improve, and control (DMAIC) cycle. In the define phase, relevant metrics are chosen, and the required data is identified. In the measure phase, the chosen sustainability metrics are measured. The improvement phase requires creating the desired future state map. The control phase facilitates the implementation of the future state map. The methodology was validated in a use case (Jamil et al., 2020).

Phuong et al. (2018) present an approach that builds upon the approach by Faulkner and Badurdeen (2014). The authors examined opportunities to integrate real-time capabilities for sustainability value stream mapping by making use of the Industry 4.0 developments. Data on the material flow is collected through radio-frequency identification (RFID) and is stored in ERP systems. These ERP systems are then used to frequently feed a dashboard that depicts relevant key figures from the sustainability value stream mapping approach (Phuong et al., 2018).

3.3. Process Mining-based sustainable value stream mapping

Lastly, the search string (‘Process Mining’ AND (sustain* OR environ* OR ecologic* OR soci*) AND ‘value stream m*’) was used to identify relevant publications in the intersection of both approaches. The search yields two results, of which only one remains relevant after scanning the title and abstract. Publications were included only if they contained other sustainability dimensions than economic aspects. The publication excluded after scanning title and abstract was the one by Nawcki et al. (2021) presented in Chapter 3.1. The abstract contained the word ‘environment’ but does not address the environment in an ecologic sense. The other paper is by Horsthofer-Rauch et al. (2022) and is presented in the following. Whilst only two publications could be found in the context of Process Mining-based digitization of value stream mapping, Scheder et al. (2023) and Phuong et al. (2018) presented approaches for sustainability-integrated value stream mapping, underlining the importance of digitizing methodology. However, both of the publications suggest approaches based on specifically recorded and stored data dedicated to facilitating a sustainability-integrated digitized value stream mapping approach. Furthermore, the approach presented by Phuong et al. (2018) only supports the digitization of the data collection and not of the model derivation.

Horsthofer-Rauch et al. (2021) analyzed the compatibility of Process Mining-based value stream mapping and sustainability-integrated value stream mapping approaches and concluded that the concepts complement each other. The authors proposed a general framework for the implementation of Process Mining (PM)-based sustainability-integrated value stream mapping (VSM) (cf. Figure 3). The framework describes the required initial as well as continuous efforts. Neither further details on required data nor a validation are provided in the publication (Horsthofer-Rauch et al., 2021).

Figure 3. Concept to create a sustainability-integrated digital value stream mapping according to Horsthofer-Rauch et al. (2021).

Overall, the publication from Horsthofer-Rauch et al. (2021) lacks details on the required data and practical implementation. Yet, the presented framework can be deployed as a basis for structuring the implementation and deployment of sustainability-integrated value stream mapping.

3.4. Conclusions from the state of the art

The enhancement of the conventional approach of value stream mapping through means of digitization or with sustainability aspects has gained great attention in research and practice. Yet, neither the combination nor an adaptable approach has been put forward that was validated in practice. The market solutions seem to be tailored approaches without the possibility of inference for other applications. A detailed analysis of the fulfillment of aspects required for Process Mining-based sustainability-integrated value stream mapping as reviewed in the state of the art is provided in Figure 4.

Figure 4. Overview of the fulfillment of relevant aspects for Process Mining-enabled sustainability-integrated value stream mapping.

Based on the identified shortcomings in literature as presented in Figure 4, the research presented in this publication aims at providing a holistic, adaptable, and generic user approach for Process Mining-based sustainability-integrated value stream mapping that supplies insights into required data, data structuring, and the technical approach.

4. Approach for Process Mining-based sustainability-integrated value stream mapping

In the following paragraphs, the general framework, required data, as well as application and validation are presented.

4.1. General framework

To structure the implementation process for Process Mining-based sustainability-integrated value stream mapping, we propose a framework adapting the work from Horsthofer-Rauch et al. (2021), which is also in line with the Cross Industry Standard Process for Data Mining (CRISP-DM). The five steps proposed are (1) defining the required scope of the analysis, (2) review and preparation of the database, (3) application of Process Mining and further analysis, (4) evaluation of the findings, and (5) deployment of the insights. Steps 1 and 2 are initial efforts that should not require repetition. The other three steps are continuous efforts, which revolve around the analysis and the interpretation of the insights. These steps should also be carried out iteratively (cf. Figure 5).

Figure 5. Framework for the implementation of Process Mining-based sustainability-integrated value stream mapping.

The implementation of a Process Mining-based sustainability-integrated value stream mapping approach requires high initial efforts but facilitates the continuous analyses of value streams with little maintenance. The steps are briefly presented in the following. Selected technical details for the implementation are presented in Chapter 4.2.

Step 1: Definition of the required scope of the analysis – The scope is defined by capturing the User Stories of the target audience of users following the Design Thinking methodology and deriving required insights and indicators from these stories.

Step 2: Review and preparation of the database – Based on a meta-model of the required data for production process modeling and indicator calculation, the available data is reviewed by an interdisciplinary team consisting of production as well as IT experts and scientists from production engineering. Subsequently, the identified relevant data is deployed through the extract-transform-load (ETL) paradigm to provide the required database for the following analysis.

Step 3: Application of Process Mining and further analysis – The mapping of the value stream is done through a Process Mining Discovery technique. Further descriptive and prescriptive performance analysis is carried out using further data analytics methods. Product family identification can also be carried out by employing Process Mining results.

Step 4: Evaluation of the findings – The insights from the analytics applications in Step 3 are examined by process experts from production to identify problematic processes and conduct further root cause analysis.

Step 5: Deployment of the insights – Insights are implemented in production and planning and control processes. To monitor improvements, Steps 3 and 4 are carried out repeatedly.

In the following, technical details are provided for Steps 2 and 3. Methods supporting the definition of the required scope in Step 1, such as the mentioned User Stories, follow the Design Thinking approach described in Chapter 2. The evaluation of insights (Step 4) and deployment of these (Step 5) can either be carried out following the guidelines for value stream analysis or are subject to company-specific procedures.

4.2. Required data

To provide a generic overview of the required data, an ontology was developed following the proposal from Knoll, Reinhart, Prüglmeier (2019). The authors suggest the use of ontologies in Process Mining projects in the context of value stream mapping to support the steps of identifying, extracting, and merging relevant data (Knoll, Reinhart, Prüglmeier, 2019). An ontology is a formalized model of a certain object and can also be understood as a meta-model of said object. It is meant to be independent of the system the data is logged in. An ontology consists of classes, relations between classes, and attributes. Ideally, ontologies serve as a communication means between different stakeholders and are readable by both humans and machines (Gruber, 1995).

The aim is to enable a Process Mining-based value stream mapping under consideration of the environmental sustainability of gate-to-gate processes in manufacturing companies. To enable this, the ontology must represent the understanding of the process perspective of gate-to-gate processes in manufacturing. Subsequently, this must be extended with the specific requirements to enable Process Mining (e.g. case ID) and the necessary data for factoring in environmental aspects. The ontology, depending on the chosen scope of the analysis, might contain aspects irrelevant to certain companies. Based upon the ontology and key performance indicators identified as relevant metrics in the user stories, members involved in the implementation of Process Mining-based Sustainability-integrated Value Stream Mapping are capable of identifying relevant data, important relationships between data, and required pre-processing.

To develop an ontology representing the production floors and manufacturing of companies in the context of value stream mapping and integrating environmental aspects, the well-established approach of Noy and McGuinness (2001) for ontology development was adopted. Existing ontologies were considered through extensive literature reviews. Only ontologies simultaneously taking into account products, processes, and resources were taken into account for reuse. Furthermore, the ontology provided by Knoll, Reinhart, Prüglmeier (2019) was deployed as the basis as its premise was closest to the considered topic of sustainability-integrated value stream mapping using Process Mining in production. Overall, the ontologies of Knoll, Reinhart, Prüglmeier (2019), Zhang et al. (2009), and Cheng et al. (2017) contributed to the new ontology. The results of the ontology development can be reviewed in the final ontology detailed in Figure 6.

Figure 6. Summarizing data ontology for Process Mining-based value stream mapping considering environmental aspects.

The ontology has four main classes Equipment, Product, Process, and Resource. The goal is to display existing relations and different aspects of such classes between or within these classes. The class Equipment holds information on manufacturing and logistics equipment used to add value on the shop floor. The class Product contains information on the theoretical build-up of a product. The class Process is made up of sub-classes of production and logistics activities that facilitate the value-add. The class Resource provides generic as well as environmentally important aspects of resources deployed in the value stream.

4.3. Technical implementation of the data base and the data analyses

To validate the scientific theory of Process Mining-based sustainability-integrated value stream mapping, the Celonis software environment was used to technically implement the concepts. The software environment leverages data from different data sources and aligns it to business outcomes. The Celonis platform enables composable business architecture, connecting data, systems, and digitization to promote and drive intelligent process execution. Data from databases, documents, and real-time event streams are integrated into the platform. After data integration, data is transformed and enriched with additional data points such as industry benchmarks and best practice models. Supported by Celonis’ patented query engine, advanced process intelligence techniques such as process analysis and monitoring, process modeling, process simulation, conformance, process predictions, and decision support can be adopted (Vogelgesang et al., 2022). The Celonis platform enables prescriptive automation, action recommendations, and prioritization and supports embeddable components in external systems. The Celonis software environment was used to develop the solution for the research project, building upon existing solutions wherever possible. The relation of the technical components required for the implementation of the developed solution can be outlined as provided in Figure 7.

Figure 7. Outline of the technical components required for working within the Celonis platform.

The process steps described in the following explain how these technical components of the Celonis platform are interlinked to supply desired analyses. Data is extracted from one or more Source Systems (1), e.g. enterprise resource planning (ERP) systems, into an SQL Data Lake (2). Here, the source tables are transformed (i.e. filtered and joined) into a form that fits the needs of the use case. The transformed tables’ relations are defined in a dedicated graph-based schema, the Data Model (3). Leveraging this data model, the tables are loaded into a purpose-built query engine, the Celonis PQL Engine (4) (Vogelgesang et al., 2022). Combined with the Process Query Language (PQL), this setting is tailored toward Process Mining capabilities and business users’ needs. The Business Definitions (5) provide a library of PQL queries. Introducing this semantical layer allows abstracting the PQL and Data Model details into reusable business-related terms. An App (6) composes a set of analytical and operational components, such as charts and tables, to uncover hidden capacities in a business’s execution. Recommendations for action might be presented as tasks for individual target persona but can also be executed automatically in the respective source system.

In the following, technical details on the Celonis-specific but customer-generic approach for the implementation of the database and the data analyses are provided.

4.3.1. Implementation of the data base

A so-called data pipeline can be implemented following the ETL paradigm as described in steps 1–3 in the previous paragraph (cf. Figure 7). Instead of extending the dimensions of the Event Log in the Event Log itself, as proposed in scientific literature, a Data Model is created to allow for a real-time update as well as the dynamic modification of the database. Provided that Source Systems and their structure differ widely between companies and even within companies, only a meta-model of the Data Model can be generically defined. Unique Data Models based on the needs of the company and the present structure of the Source System must thus be defined during the implementation of the approach. An overview of the necessary steps to build up the required Data Model is presented in the Figure 8.

Figure 8. Overview of the process of identifying and pre-processing data.

Leveraging this Data Model, the tables created and connected within that Data Model are loaded into a purpose-built query engine. The query engine is the service through which applications enable users to efficiently roll out analytics on large datasets. Details on the implementation are provided in the following Chapter 5.3.2.

4.3.2. General application of the data analysis

Before the implementation of the value stream analysis and the product family formation are explained in detail, a brief overview of the approach for implementing analyses in the Celonis platform is provided.

As described in the preceding chapter, the data model is loaded into a purpose-built query engine called Celonis PQL Engine. This enables user-friendly access to the database. This setting is tailored toward Process Mining capabilities and business users’ needs (Vogelgesang et al., 2022).

The so-called Business Definitions provide a library of PQL queries and the respective business context. Introducing the Business Definitions allows for abstracting the PQL and Data Model details into reusable business-related terms and, thus, the scalable productization of business knowledge. The following exemplary PQL query returns the average throughput time in hours across all production order items aggregated on produced materials:

PU_AVG(MATERIALS,CALC_THROUGHPUT(CASE_START TO CASE_END,REMAP_TIMESTAMPS(‘PRODUCTION_EVENTS’.‘EVENTTIME’, HOURS)))

An App composes a set of analytical and operational components, such as charts and tables, to uncover hidden capacities in a business’s execution. The procedure for the application of the aforementioned aspects and the implementation of analyses can be seen in Figure 9.

Figure 9. Overview of the procedure for enabling the analysis.

Using the scheme depicted in Figure 10, the value stream analysis was implemented, and the product family formation was integrated. Details on the implementation and integration of these are presented in the following.

Figure 10. Schematic overview of the functionality of the Process Explorer for the value stream map.

The value stream analysis using Process Mining is realized through readily available functionalities for Discovery and performance analysis on the Celonis platform. The value stream map is presented through a directly-follows graph in the platform’s Process Explorer, which fulfills the purpose of the Discovery technique. Results from the application of the Process Explorer on real production data from the use case are presented in Figure 10.

Performance analysis through further value stream-related key performance indicators (e.g. scrap), which are implemented as Business Definitions and presented through different dashboards readily available in various Apps on the Celonis platform. Depending on the thresholds individually defined by the company, prescriptive information (e.g. on component prioritization) can be provided through further Apps. Providing the value stream map via the Process Explorer requires little individual adjustment and can be determined as an ‘out-of-the-box’ solution. Some of the general value stream key performance indicators, especially those related to time and scrap, are readily available in the form of Business Definitions in the Celonis platform. Other key performance indicators have to be developed individually (e.g. supplier reliability) if company-specific calculations deviate from common formulas or if required data is not stored in standard tables of common IT systems. To drive tangible value, persona-specific dashboards are created to lead users efficiently in their analysis. These dashboards should be created based on the individual needs within a company and can be realized in a low-code and no-code environment on the platform. Details on the analyses developed within the research project based on the needs of the application partner can be found in Chapter 5.4.

Part of the value stream mapping methodology is product family formation (cf. Chapter 3.1). Within the conventional approach, product family formation supports the identification of product families from which a representative product is chosen. This representative product is then the basis for value stream mapping efforts. Insights from the value stream map of this product are further used to derive insights for all other products of the common product family. Optimization is usually carried out undifferentiated for all products in a product family. While the presented digitized approach actually does not necessitate the forming of product families, as every product can be analyzed at no additional cost, product family formation is still helpful for further optimization efforts (Horsthofer-Rauch et al., 2022). Both apriori and a-posteriori approaches were researched, reviewed, and validated. Approaches apriori Process Mining were reviewed but did not yield meaningful results at an acceptable computational time. Thus, approaches a-posteriori Process Mining were researched. After review, the approach provided by Meincheim et al. (2017) was chosen. Meincheim et al. (2017) present a trace clustering approach, which divides cases into different groups based on the similarity of the individual process models. According to a threshold, either a new cluster is created for the trace or it is added to an existing cluster (Meincheim et al. (2017).

We implemented the approach using the programming language Python, deploying the libraries NumPy, Pandas, and PyCelonis. The latter allows frictionless integration between the Python-based code and the Celonis platform. The adapted approach requires first the conversion of information in the event logs into traces, i.e. sequences of activities of a case. After the traces are available, trace profiles are built using one-hot-encoding to see which activity is represented in which cases. Furthermore, the sequence of activities is analyzed through one-hot encoding. This builds the basis for the clustering, which was realized through a k-Means clustering algorithm. Finally, the cases in the clusters are reviewed for unique material numbers within each cluster. A cluster then represents the product family, whilst the associated products are identified by the material numbers. As the k-Means algorithm requires a prior determination of the number of desired clusters, the solution within the Celonis platform allows for choosing between different numbers of clusters.

The implementation within the Celonis platform is depicted in Figure 11.

Figure 11. Overview of the product family clustering review dashboard.

4.4. Practical application and validation

In the following paragraphs, the practical application and validation of the methodology and the technical implementation are carried out at the HAWE Hydraulics SE. The company is a manufacturing company specializing in hydraulic aggregates and components. Its product portfolio contains over 200 standard products and further customized products with varying production volumes. Group production, as well as workshop production, is available. The value stream of a product often extends across different plants, partly due to the high vertical integration of manufacturing steps. The company used conventional value stream mapping regularly before the research project and has involved production experts and supply chain experts in the process. Due to the high product variance and the physical distance between the individual production sites, the efforts for value stream mapping through the conventional manual approach are high. Furthermore, product family formation based on standard processes also requires inefficiently high efforts. To ensure the protection of sensitive information, details on the specific data values are not provided.

For the application, the methodology presented in Chapter 4.1 was deployed. To capture the needs of the users, several workshops were carried out iteratively involving stakeholders from various production-related departments following the Design Thinking methodology to identify relevant Personas and define their User Stories. Based on these User Stories, relevant key performance indicators and information were derived, and Business Definitions were created. Subsequently, the database was created, connecting relevant systems and data points to the Celonis platform. Relevant data was identified using the ontology and deriving information needs from the User Stories. Eventually, the data analyses supporting functionalities based on user needs were assigned and implemented by Celonis developers. The relationships between the User Stories, relevant key performance indicators, and functionalities are provided in Table 1. Under standard value stream indicators all economically focused value stream indicators of relevance to the company are summarized. This encompasses indicators such as lead time, process time, stock range, etc., depending on the company’s preferences.

Table 1. Overview of User Stories used for the development of the solution.

User Story		Required Key Performance Indicators/ Relevant Information	Implementation
1	I need to analyze the value stream map, incl. material flow, information flow, and indicators, so that I can find issues within the actual state.	standard value stream indicators; value stream map	all standard value stream indicators implemented; visualization of the value stream in the Celonis Process Explorer; drill-downs/detailed views for work stations; information flow not displayable due to missing systematic digital information
2	I need to find waste within the value stream.	information on transport, inventory, motion, waiting times, over-production, over-engineering, and defects	filtering and sorting mechanisms for identified indicators; Sankey diagrams on transportation between sites within the same value stream; dashboard for quality issues; no consideration of motion and over-engineering due to missing data/data granularity
3	I need to examine process steps regarding times, personnel, machinery, work in progress, and stock within my production processes.	circulating stock; cycle time; information on resources and stock	all defined indicators implemented; drill-downs/detailed views for work stations; dashboard for quality issues
4	I need to assess overall utilization and bottlenecks so that I can re-plan production programs.	utilization rate; capacity	all defined indicators implemented; Sankey diagram on capacity requirements (work station drill-down)
5	I need to establish transparency of the interdependencies in the value stream with regard to capacity.	information on value stream families using the same workstations	all relevant indicators implemented; Sankey diagram on capacity requirements (work station drill-down)
6	I need to establish a flow between departments as well as between work centers to optimize the value stream holistically.	standard value stream indicators	all standard value stream indicators implemented; Sankey diagram on capacity requirements (work station drill-down)
7	I need to ensure that no unnecessary stock and waiting times or over- and undercapacities occur between and within supply chain steps.	circulating stock; cycle time; information on capacity requirements	all defined indicators implemented; overview on inventory and component availability; prescriptive analysis on component prioritization; Sankey diagram on capacity requirements (work station drill-down)
8	I need to see possible alternative resources so that I can assess the best strategy to deal with restrictions.	information on alternative work plans and technology redundancy	depiction of alternative work plans
9	I need to see an overview of all orders in the value streams aggregated by product type or value stream group with drill-through possibilities for detailed analysis.	information on products and corresponding components	implemented through a Bill of Material explorer, allowing filtering and drill-down in all other analyses
10	I need to see supplier performance regarding lead times and punctuality.	supplier lead times; supplier reliability	all identified indicators implemented
11	I need to see power consumption and related CO2-equivalents to identify process steps and products that create high emissions.	power consumption; CO2-equivalents	all identified indicators implemented

Three examples of the implementation of the User Stories are depicted in the following paragraphs to provide a more detailed view. The three examples aim to provide a broad spectrum of implemented functionalities.

4.4.1. User Story 1

The first user story required the depiction of information relevant to value stream analysis, such as the value stream map, as well as relevant key performance indicators. From a user perspective, support for in-depth analysis of a certain material number is demanded. The event log and the connection of different abstraction levels for resources and activities available in different ERP system tables mainly facilitate the analyses. Due to the static nature of pictures, the dynamic filtering and deep-dive options cannot be holistically depicted. Figure 10 (cf. Chapter 4.3.2) provides an overview of the dashboard provided for this use case.

On the left side of the dashboard, the directly-follows graph depicts a simplified version of the value stream map. Above the graph are the number of closed and open production orders of the products analyzed, the number of materials (i.e. products), the number of involved workstations, and the number of production operations. The sliders next to the graph allow the opening of more connections and activities until all connections and all activities the material has run through in the past are depicted. The connections between activities contain information on throughput times (mean and average) and the number of occurrences of a certain connection. The selection breakdown on the right allows changing dimensions and filtering. Here, the process explorer dimensions (i.e. depth of the information provided on the graph node) can be chosen, and filters for the order status, the connection, the organization (i.e. sites), and material information can be applied. Filters applied here can be carried out throughout the following analyses.

4.4.2. User Story 7

For this use case, stock and waiting times should be provided to help keep inventory and respective costs low, establish flow, and see process inefficiencies quicker. From a user perspective, a starting point to find optimization potential in all value streams and review certain value streams or product families in more detail is demanded. The required key performance indicators and further information were set up in different dashboard views. Regarding the data, inventory is represented through the physical stock recorded in respective IT systems and the designated and recorded value per part. Figure 12 shows the inventory classification view providing information on, e.g. understock, overstock, and active stock.

Figure 12. Overview of the dashboard for the inventory classification view for User Story 7.

The inventory classification view allows the analysis of inventory with regard to its quality, possible excess or understock, and value. The view can be filtered by country, site, material group, and material number. In the primary view, a graph depicts the inventory classification in terms of the number of materials of a class (e.g. understock or overstock) and the value. Following the tabs of the same field, the actual days-on-cover compared to the target days-of-cover, the safety stock, or the actual lead time can be reviewed. Down below, action tables are provided to guide the user in finding issues on different levels (country, plant, and material).

4.4.3. User Story 11

This user story focuses on the ecological aspects of the value stream. From a user perspective, support for the identification of ecologically intense production steps, resources, or materials is demanded. Thus, in this view, power consumption, scrap, transport, and CO₂-equivalents are visualized. This enables optimization as well as reporting of ecologic sustainability aspects. Power consumption represents the power consumed during a production process step for a single material on a single machine or an aggregation of this, depending on the chosen filters. Scrap refers to produced scrap of a certain material. The transport shows the movement of parts between production sites. The CO₂-equivalents were calculated using a factor representing the power mix available at the production site. The dashboard created for this user story is depicted in Figure 13.

Figure 13. Overview of the sustainability dashboard for User Story 11.

On the left side on top, the number of materials, total scrap, number of materials, total CO₂-equivalents, and power consumption are represented. A histogram allows for analyzing the ecological aspects over time. To the right of the histogram, a Sankey diagram shows the transport of material from one production site to another. A table on the bottom lists all values for the ecological aspects of materials, resources, or activities. The selection breakdown on the right sums up the represented materials, workstations, and production orders. The dashboard can be filtered by operation or material information. The table can be sorted from high to low or vice versa for each aspect.

5. Discussion and outlook

The results of the research, as presented in this paper, provide answers to the posted research questions and show that the main research goal could be reached.

Through our research and development activities, we were able to implement theoretical concepts for Process Mining-based sustainability-integrated value stream mapping on real live data from production. Conventional value stream mapping cannot only be digitized but also extended by further dimensions, i.e. sustainability, and further analyses, e.g. prescription. A generic overview of required data was provided in the form of an ontology, enabling potential other interested parties to identify and pre-process relevant data (Research Question 1). The Process Mining type Discovery and respective performance analyses support the value stream analysis, while a Trace Clustering approach can be implemented to facilitate the product family formation (Research Question 2). Available apps on the Celonis platform allow the implementation of dashboards that provide value stream maps as well as other relevant indicators and information in an easily accessible way (Research Question 3). The depiction of the actual value stream map through a directly-follows graph differs vastly from the conventional value stream mapping notation. However, through user-friendly dashboards, all relevant information can be depicted accessibly. The software environment allows real-time analyses for each individual product or order and broadens the view of the value stream to support various production-related tasks. In the context of prescription, suggestions on component prioritization to improve stock levels were provided.

Overall, we were able to demonstrate the potential and the efficiency of a digitized version of value stream mapping with a use case on real data. However, as production environments tend to have historically grown key performance indicators and databases, the developed solution is not fully transferable and scalable.

5.1. Benefits of the developed solution

The developed solution is tailored for the use case of the HAWE Hydraulics SE but can be adapted for other manufacturing companies as well. The solution allows for real-time analyses for all product value streams efficiently. Compared to the conventional value stream mapping methodology, this facilitates analyses much more frequently and comprehensively. In high-variance, high-volume production environments, this increases the economic efficiency of value stream mapping for various products or product families. The abstraction of knowledge through the data model and Business Definitions enables efficient insights into various aspects of production (e.g. cycle times, resource consumption, etc.) and the supply chain (e.g. supplier reliability, customer order behavior, etc.). Additionally, the use of aggregated historical data retrieved from various databases facilitates more objective and effective performance analysis of value streams compared to the snapshot data used in conventional value stream mapping. The data integration is system-agnostic and dynamically modifiable. These aspects enable later extensions and ensure comprehensibility. Furthermore, data from upstream and downstream supply chain activities can be integrated to extend the view on factors influencing value stream performance.

5.2. Limits of the developed solution

Data acquisition in production tends to be less systematized, causing issues regarding the quality and granularity of recorded data. In some areas of production, manual confirmations of production orders are still required. These manual confirmations are prone to error and may provide a less realistic view than automatically recorded data. Especially data on social or ecological sustainability has just recently been acknowledged as relevant data in production. Thus, data regarding these dimensions are scarce compared to data on time, quality, and costs in production.

Additionally, the use of predictive and prescriptive analyses is possible but requires high-quality data and the willingness to incorporate such data in planning activities. Detailed prescriptive analyses supporting production planning and scheduling require extensive simulation capabilities, which were not considered in the presented research project.

Overall, the initiation of digitized sustainability-integrated value stream mapping requires high effort and access to detailed process knowledge. An interdisciplinary team consisting of IT experts, data experts, and production experts is needed to set up the solution initially. Depending on the state of the available data and required key performance indicators, custom-fit transformations and Business Definitions are needed. A detailed cost-benefit analysis must be carried out to ensure a lasting value-add for different departments.

The validation of the methodology is also limited to an implementation using the Celonis EMS. Here, the pre-defined algorithms for the different Process Mining types and data analytics capabilities were used. Thus, the benefits of different techniques for implementing the different types of Process Mining are yet to be assessed.

Furthermore, concepts for automated value stream design based on the digitized value stream analysis were not considered and must be subject to further research.

5.3. Future research

To support and enhance digitized sustainability-integrated value stream mapping, the following aspects have to be further researched. First, the database should be reviewed and optimized with regard to recording and storing production data systematically to grant a holistic, representative view of production. Secondly, the topic of OCPM can enable an even more holistic view with regard to upstream and downstream operations and end-to-end order processing. This helps streamline organizational processes overall and simplifies reporting of relevant (sustainability) indicators. Furthermore, the integration of predictive analysis may amplify the value-add of digitized value stream mapping, especially regarding operational topics. Adding to this topic, simulations based on value stream data can facilitate prescriptive analyses and help create a more effective and efficient way of value stream design. Overall, the assessment of different techniques for the different Process Mining types should be done in future research to see if specific techniques produce better outcomes with regard to value stream mapping. Finally, the aspect of the interactions between the user and the software environment must be further studied to understand the possible shortcomings of the solution.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research project was funded by the Bavarian Ministry of Economic Affairs, Regional Development and Energy [grant number DIK-2004-0018// DIK0126/01].