Industry 4.0 Guide

Ultimate list of Process Mining case studies in manufacturing

List of Process Mining cases studies in manufacturing. Overview of application types and use cases in different production settings.
Ultimate list of Process Mining case studies in manufacturing

Why Process Mining in manufacturing?

Process Mining is an emerging technology that leverages event logs from IT systems to provide an understanding of the as-is processes. As data transparency is a prerequisite for the digital transformation in manufacturing, Process Mining is a key enabler to create a data-driven shop floor. The effects can be dramatic and typically include higher throughput, improved machine utilization and reduced non-conformance costs. With Process Mining, production bottlenecks can be recognized and unnecessary process steps can be reduced. Process Mining allows manufacturing companies to identify process inefficiencies, understand their root causes and continuously improve production processes based on data-driven insights. The three main tasks performed by Process Mining software are process discovery, conformance checking and process model enhancement. To learn more about what Process Mining is and how it works, feel free to read our introductory article "Process Mining for Manufacturing".

Benefits of Process Mining in manufacturing

With Process Mining, Manufacturing companies get full transparency about the status of their production. The resulting benefits are vast and include: 

  • Improved OEE and machine utilization: e.g. by improving the reactive maintenance rate and turning reactive maintenance into predictive maintenance. By analyzing historical data, optimal maintenance windows can be identified to reduce the time spent on reactive maintenance.
  • Reduced quality costs: by identifying correlations between material defects and rework activities, quality processes can be automatically improved resulting in reduced costs of quality.
  • Improved throughput: by identifying unnecessary waiting times, process bottlenecks can be removed resulting in reduced cycle times and higher throughput.
  • Improved schedule adherence: by improving process conformance, the late completion rate can be reduced enabling high priority orders to be shipped in time.

Process Mining case studies in manufacturing

Typical application areas in different production settings

To get a comprehensive overview about the usage of Process Mining in manufacturing, we have conducted a systematic literature review of research publications about Process Mining in industrial settings. Process Mining in manufacturing is an emerging discipline, and the number of publications with application examples is constantly growing. We haved used the following publications as a starting point for our research:

Dreher et al. use the SCOR-model to cluster the available case studies along the process types Plan, Source, Make, Deliver, Return and Enable. The Make process is the core process for production including all activities that transform material to finished products. Accordingly, the most amount of publications from the study were assigned to the Make process (13 publications in total). With our research, we found additional publications related to the Make process and have analyzed 22 case study publications in total. Dependent on the production type and industry, the expected benefits and improvement opportunities of Process Mining are different, which led us to a clustering according to the production type. As the level of use case details and application domains provided by the authors vary, we have chosen the rather generic and high-level production type categories  Job shop / Low volume production, Batch / Serial / Mass production and Continuous production. In total, we identified 4 case studies in the category Job shop / Low volume production, 15 case studies in the category Batch / Serial / Mass production and 3 case studies in the category Continuous production (Table 1).

Table 1: Systematic literature review: 22 case studies about Process Mining in Manufacturing


In addition, we analyzed the case studies according to the three main tasks performed by Process Mining software Process discovery, Conformance checking and Process enhancement and identified typical goals of the Process Mining activities per production type category.

In the Batch / Serial / Mass production category, most case studies focussed on Process discovery (86%) and Process enhancement (53%). The majority of case studies were conducted in the Automotive sector. Typical goals included the detection of bottlenecks and outliers, the identification of anomalies and root causes of poor quality. The overall targets for process enhancements included the reduction of lead times, the optimization of OEE, the ability to implement traceability and gain production flexibility.

In the Continuous production category, all case studies targeted conformance checking. The business goals included the monitoring of workflow conformance to identify operational problems associated with human decisions and ensuring safety and procedural compliance to reduce non-conformance costs. 

In the Job-shop / Low volume production category, all case studies focussed on process enhancement. Here, examples for business targets include the analysis of workload in make-to-order manufacturing characterized by very long lead times and the rearrangement of process steps to reduce delays.

Examples of Process Mining in manufacturing

In this section, we will summarize case study examples to provide a better understanding of application areas, challenges and benefits for Process Mining in manufacturing.

We focus on the following selected case studies, in case you are interested to discuss additional examples, please just let us know:

Want to get access to the full case study list? Just let us know and we will contact you soon.


Case study: BMW Process Mining

In 2017, the automotive manufacturer BMW Group introduced Process Mining into their production processes with the aim to enable best production with the highest quality possible. Prior to this, production errors and their root causes were analyzed through direct observations, requiring long-term experiences, and testing. More information from sensors was already available, but not used to monitor entire processes. BMW's IT landscape was fortunately similar in each plant, the understanding of a process in the production environment was however rather difficult to reach based on the collected data only. One sub process in manufacturing could easily have more than 1,000 sensors collecting data. This differs from all non-production processes, such as procurement processes. The initial application of process mining at BMW was the optimization of a new paint shop, but soon the usage was extended along the value chain. Current plans include further scaling and extension with complementing technologies like RPA and AI.

Process Mining along the value chain

What were the benefits of implementing Process Mining?

  • Increased transparency: The as-is process, which gets executed, is also the basis to understand interdependencies and their complexity and is essential to align communication. Connecting quality and machine data with stepwise process data supports root cause analysis enormously. Also, connecting cost components the exact costs per produced vehicle is calculable, with transparency of e.g. colours that are more expensive due to higher energy costs as others
  • Increased agility: Process Mining created a digital twin of the production and allows accelerated issue detection and reaction time, achieving more agility. For example, process conformance checks reveal unplanned discrepancies between the as-is and design model, and is necessary to define measurements to adjust workflows
  • Improved productivity and reduced costs: Near-real-time connections to different systems allow to use Process Mining as a monitoring tool for decision support. With this, productivity and quality can be increased, rework and production costs decreased. For example, the identification of bottlenecks is effortless possible, also the direct evaluation of improved process variants can simplify the way towards process enhancement
  • Potential for continuous improvements: (Re-)Assessment of defined KPIs as well as their measurements to reach them can lead to further performance improvement when looking into process details. Expanding Process Mining into other plants enables a ecosystem for comparison and learning potential

Case study: Process Mining at e.GO  

e.Go was founded in 2015 with the aim to produce electric vehicles for short-distance routes. The production is semi-automated and consists of human staff at the stations and Autonomous Guided Vehicles carrying the cars through the production process automatically, without any buffering in between. e.Go applied process mining to improve its production, based on a dataset that includes recordings of every production step. 

What were the benefits of implementing Process Mining?

  • Bottleneck identification: Service and waiting times between stations were visualized with the as-realized model, showing the dependency within the sequence and thus increased waiting times at prior stations. The idle time thus results in a slightly longer service time at a subsequent station. Moreover, main stations that depend on sub-assembly lines influence all main stations negatively. 

Case study: Process Mining at Geberit

Geberit is a Swiss producer of sanitary products distributed in 50 countries and is considered as leading in its market. For the process mining use case, the factory in Switzerland which produces a plastic actuator with 18 different variants was chosen. The production process consists of molding, assembling (1&2), sorting, and packing, and was semi-automated. The unique ID is the serial number and gets scanned through DMC readers at the machines and stored with the scanning timestamp. The collected dataset therefore already reached 700,000 event logs.

Automatic data collection for Process Mining


What were the benefits of implementing Process Mining?

  • Identify non-conformances: With the as-realised model, Geberit got proof about the inconformacy of some process flows to the originally designed ones. For instance, the as-realised model showed loops and reverse steps in the process flow. Reasons for that were unplanned rework processes and poor data quality, resulting in a false sequence.  
  • Identify bottlenecks: Geberit also defined segments based on the time between two DMC logs. Taking them in relation to the overall throughput time, the segment which took the longest on average was detected easily: The segment where sorting and batching into mobile racks was performed automatically, and an operator loaded the packaging line afterwards. The packaging line with the next DMC reader showed for some times an unnecessary idle of 59 minutes in total. With further investigations including the domain experts,  three root causes got identified. First, due to the shift change the material was on-hold, until the next operator began his/her shift. Second, unknown inefficiencies in a product variant change were detected. Third, completely filled racks were not perceived immediately every time.
  • Remove bottlenecks: Based on the identified bottlenecks, Geberit improved the as-realized process. To minimize the detected idle time at the packaging line, the loading should happen directly before the shift. Moreover, the scheduling got adjusted so the changeover times between variants were as short as possible. Lastly, the loading of the packaging got automated. 

Lessons learned: Process Mining projects

Our research shows that all manufacturing companies that implemented Process Mining were able to realize substantial business benefits. But as Process Mining is a relatively new technology, shortcomings and challenges for the implementation of the technology still exist. We have summarized a few project management lessons learned from the publications below. 

  • Keep a future- and solution oriented perspective: Process Mining detects inefficiencies that should not be used for finger-pointing. The only target is to improve the process for the future
  • Change Management: Explaining the technology works best, when demonstrating a demo with actual data from the business unit. This showcases the potential best and improves further communication
  • Quantifying business value: Quantifying time savings, improved quality, or expected cost reductions as pure consequence of Process Mining are not always straightforward. Re-investing saved resources is needed to support and reward business units achieving their business targets.
  • Focus on actions: Data analytics are not influencing any business value if no action is executed. Often the first focus lies on monitoring processes. Optimizations are rarely included in the business KPIs which decreases the engagement for those activities.
  • Preparation is crucial for scaling and rollout phases: Trained key users and other support is required to successfully scale and rollout Process Mining into the production plant
  • Different support systems: Depending on the people's engagement, motivation, working atmosphere and more, one plant may differ from another. Some are more independent and drive their own use cases, some need more consultancy and guiding. Thus, a Center of Excellence is very valuable to provide supporting services dependent on the needs of the staff.

Data quality is a major challenge for Process Mining in production

One of the particular challenges for Process Mining in production environments is the quality of the event logs. Very often, production processes are unstructured and non-linear, involving a high degree of manual activities. IT-system landscapes tend to be fragmented and the integration of heterogeneous data sources is challenging. We have identified the following challenges related to data quality:

  • Poor data quality from automatically collected data: Automatically collected data very often stems from heterogeneous operating equipment, e.g. machines and sensors from different vendors. As automatically collected data may be wrong or out of sync, it is important to detect incorrect data immediately, before including it into further analysis. Furthermore, the required data might not yet be captured from the physical process. It is costly to replace older machines that have limited sensory capabilities. Hence, many manufacturers are incrementally introducing automated data capture by retrofitting existing production lines. But sometimes not all machines are covered and the dataset is therefore incomplete. This might limit the scope of analysis and improvement actions that can be derived from process mining.
  • Incomplete data from manual workflows: In many production environments, the share of human activities as a crucial part of a production process is still very high. However, the manual activities and the resulting data are often not adequately captured. Collecting data manually with paper lists or at PC stations that are not in proximity to the point of value generation is difficult. This usually results in poor data quality. For example, some process-related data is not recorded (e.g. activity start times) or is only partially recorded (e.g. information about employees involved in the process). Many publications refer to situations where the timestamp of the start and end point of activities at the station were not captured precisely by the operators. Or rework operation events were modified and caused a deviation of actual starting time and recorded starting time. Thus, the sequence of rework steps were not reliable any more. Accurate capture of all process execution data would allow better Process Mining results, e.g. by also considering activity durations or analyse behaviour of employees handling the process.
  • High process variability: For many manual jobs, it is difficult to predict the processing time of each activity. Even if activities seem to be similar, the processing time of an activity varies according to the level of difficulty and to the operator performing the activity. Most IT production systems do not cover data about process difficulties or data about the required competencies. Therefore, reflecting the level of task difficulty and skills needed in event logs would be beneficial to improve process analysis. In environments with high process variability, not only activity relation and frequency data is needed but also the affected resources and detailed parallel relations should be covered in the Process Mining event logs.
  • Missing links between data: Due to a fragmented IT landscape, data sources are often different, data is stored in different formats and linkage between data is missing. Thus,  the extracted process models are often not accurate. 
  • Duplicates: sometimes production processes are non-linear and include loops and redundancies. For example, rework procedures or sub-tasks performed at substations make result in repeating data collection procedures. If IT-systems are not properly prepared, such routines may result in data duplicates that negatively impact the quality of Process Mining.  

To be able to get valuable insights from Process Mining in production environments, it is important to focus on data quality. On of the key drivers for better data quality in Process Mining is to include as much data from manual activities as possible. A solution such as the WORKERBASE Connected Worker Platform extends Process Mining software with event logs from manual activities and can thus drastically improve data quality and overall results. If you are interested in more information about how companies established Process Mining in production environments, please do not hesitate to contact us!

Authors:

Sabrina Joos, Solution Consultant WORKERBASE

Thorsten Krüger, Co-Founder WORKERBASE

Request a live demo

Would you like to learn more about our solution?
Fill in the form and we will contact you to schedule a live demo.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.