With recent advances in IT infrastructure in manufacturing environments, a large amount of production data is collected and stored in a database at various stages of production. Therefore, in this thesis, an approach based on process mining is proposed for analyzing complex manufacturing processes. Process mining is a useful tool for acquiring process-related knowledge as it enables users to extract not only production process models, but also several performance metrics related to processes, resources, and tasks.
Furthermore, a case study is conducted to support the proposed framework with an event log of an electronic component manufacturing process.
Introduction
Motivation and Background
Process mining is able to extract process-oriented knowledge from event logs (Van der Aalst et al. 2004;). Process mining offers many analysis techniques, and some of them can be useful for the analysis of production processes. First, process mining has various mining tools that discover a production process model that provides insight into actual production processes.
Objective
In addition, we used several performance analysis techniques to monitor current resource operations in terms of their idle time. The purpose of this thesis is to monitor and diagnose the production capacity for each task in terms of their utilization rate and preparation time. The rate of return of production is important to reduce their lost rate, which is related to production costs.
Thus, it is essential to examine their yield rate, work time, and wait/turnaround time for each task.
Outline of the thesis
This analysis includes structuring the relationship between resources and tasks with respect to their work frequency, and analyzing machine idle time and work time. Machine allocation analysis will use organizational miner to structure "who or what" resource worked for an activity, and to examine resource allocation in relation to work frequency. Various literature research has suggested many different performance indicators such as cycle time, throughput, return rate and lost rate.
Related Works
Manufacturing Process Management with Data Analysis
- Simulation for Manufacturing Data Analysis
- Data mining for Manufacturing Data Analysis
For example, Chen et al (2004) used association rules for defect detection by determining the association between different machines and their combination with defect (Chen et al. 2004). According to the research on the review of application of data mining in manufacturing, data mining techniques are useful in improving the capability of manufacturing enterprises by providing useful knowledge for better decisions. Harding (2006) addressed that the knowledge discovered with the help of data mining is sometimes too complex to understand, therefore they must make an effort to improve the expressiveness of the knowledge.
To address the limitation, this thesis will use process mining techniques to analyze manufacturing data analysis as process mining is useful for visualization.
Process Mining
- Overview of Process Mining
- ProM framework
- Application of Process Mining
ProM supports many process mining techniques in the form of plugins and uses an MXML file as input format. Traditionally, process mining has focused on the control-flow perspective (Van der Aalst et al.). There is an increase in the application of process mining in real environment industries with regard to the different perspectives.
Overall Funding Methodology for a Versatile Analysis of Real Event Logs Based on Process Mining.
MES (Manufacturing Execution System)
A Process Mining Framework for Manufacturing Process Analysis
Problem statements
As batches move through the production processes, event logs are automatically collected on the MES. Therefore, it is possible to generate a social network model and structure their relationships on the transfer of the works. First, we will structure the relationship between resources and tasks to reduce failures and improve resource performance.
Second, performance analysis of each resource against idle time is needed to know how long the resources have been waiting. Therefore, it is important to reduce the production cost by reducing the loss rate and total time. Analyzing the performance of each task can provide information on production time and lost rate.
Compliance of standard model To compare the standard model with event logs by calculating fitness. To visualize bottleneck point on the process model and compare the bottleneck points between different production patterns. To structure relationship between machine and task and compare working frequency of machines for each task.
A Process Mining Framework
- Data preparation
- Data preprocessing
A manufacturing process consists of the sequence of activities to make final products from raw materials. In the manufacturing processes, it is possible to record the event type that identifies the transactional information at a given time. The time (i.e. timestamp) of the event is to show the sequence of events in a case.
Looking at the output event log data in Table 4, we can formalize the event log as follows. An event log, denoted L, consists of a set of process instances or instances, and each instance is described by a sequence of events. C is a set of events for its single instance where a collection of all instances is an event register (L).
Examples always have a trace denoted as σ∊E*. C=E* is the set of possible sequences of events (traces that describe an instance). For example, πA(e) is the activity associated with event e, πY(e) is the transaction type associated with event e, πR(e) is the resources associated with event e, πT(e) is the timestamp associated with event e, πC(c) is the trace associated with instance c, and πD(e) is the quantity data associated with event e. After the data cleaning step, the data conversion step converts the data into a suitable format for process mining, such as MXML (Mining eXensible Markup Language) and XES.
Since the process mining tool, ProM, requires MXML as input file format, data conversion is necessary. The number at the beginning of the line identifies the LotID (ie a product lot) being executed.
Manufacturing Process Mining and Analysis
- Process perspective: Manufacturing Process Visualization
- Resource perspective Ⅰ : Machine-to-Machine interrelationship Analysis
- Resource perspective Ⅱ : Machine Utilization
- Task perspective: Monitoring & Diagnosis of task performance
A production process model is built based on an event log, so it captures the actual behaviors seen in the event log. In the case of fuzzy mining, it has advantages in finding the abstraction process model when the process is complex (i.e. has a large number of activities). 5 shows an example of the propagation of events in the log according to the current time and is sorted in descending order based on the duration.
It shows that using the dot plot can clearly describe the real world situation based on the information in the log. The goodness-of-fit measure is a discrepancy value as a result of replaying the log in the model. The replay of a record starts with marking the starting place in the model and then the transitions belonging to the recorded events in the trace are activated one after the other (Rozinat and van der Aalst. 2006).
It also counts the number of tokens left in the model, indicating that the process is not completing properly (remaining tokens). Second, the manufacturing process most likely has a sequential flow that does not include any parallelism in the production flow. In the case of activity, the working and waiting time of two activities can be identified.
And for the machine working time/idle analysis, we used the “Basic Performance Analysis” and the “Dotted Chart” to delineate other performance indicators such as time and equipment efficiency in the logbook. For a task, the working and waiting/transport time of two activities can be identified. In combination with the pattern analysis from the previous section, the process mining tool can enable working time and waiting/transport time analyzes between certain activities.
In order to improve the utilization rate of the production process, it is necessary to consider the amount of product in the analysis.
Case Study: An Electronic Components Manufacturing Process
Context
Summary of manufacturing process mining and analysis
- Manufacturing process visualization of process perspective
- MTM interrelationship analysis of resource perspective
- Machine utilization of resource perspective
- Monitoring & Diagnosis of task perspective
A dot is colored based on the name of the activity and the cases are sorted by case duration. The gap between two dots describes the time interval of the two events, and the longer distance affects total lead time. First, conformance checking provides a fitness value of the discovered process model from event logs by comparing with the event log.
Bottleneck analysis visualizes the bottlenecks where an activity has a long time in the discovered process model. A social network shows how lots are transferred between machines, and the table shows which machine has the most frequent relationship between two other machines. According to the result, machine M0052 has the highest betweenness value, and the machine is in the important location of the network.
One of the main goals of production management is to increase production efficiency by reducing overall costs and lead time. For this reason, root cause analysis is necessary to find the activity with a high defect fraction and lead time. Production time analysis focuses on high lead time, while production quantity loss focuses on defective fractions.
Actually, we have already introduced another root cause analysis, bottleneck analysis which detects high lead time tasks in the process perspective. The proposed bottleneck analysis finds a bottleneck with a high lead time on the manufacturing processes.
Conclusion
34;Making Compliance Measures Actionable: A New Compliance Analysis Approach," In Proceedings of International Conference Business Process Management Workshops, Springer Berlin Heidelberg. Data Mining-Driven Manufacturing Process Optimization," In Proceedings of International Conference on the World Congress on Engineering (WCE) 34 ;Using Simulation in Manufacturing and Logistics System Planning,” In Proceedings of the VTT Manufacturing Technology.
Digital Models for Manufacturing Process Visualization,” In Proceedings of the International Conference on Integrated Logistics, Singapore, 113-122. 34; Application of Process Mining in Healthcare - A Case Study in a Dutch Hospital," In Proceedings of the International Joint Conference on Biomedical Engineering Systems and Technologies, Springer Berlin Heidelberg, CCIS. 34; On the Performance of Work Processes with Dispersed Actors: Does Place Matter?" In the Proceedings of the 5th International Business Process Management Conference.
34;Comparative Advantage by Manufacturing Execution Systems," In Proceedings of Advanced Semiconductor Manufacturing Conference and Workshop, ASMC 96, IEEE Cambridge, 179-184. 34;Process Mining: Overview and Prospects of Petri Net Discovery Algorithms," In Proceedings on 26th international conference on Transactions on Petri Nets and Other Models of Confluence II, Springer Berlin Heidelberg. 34;Review on Application Data Mining in Product Design and Manufacturing," in Proceedings of the Fourth International Conference on Fuzzy Systems and Knowledge Discovery - FSKD 2007, IEEE Los Alamitos.
34; Process Mining: Discovering Workflow Models from Event-Based Data," in Proceedings of the 13th Belgium-Netherlands Conference on Artificial Intelligence (BNAIC. 34; Discovering Implicit Dependencies between Tasks from Event Logs," in Proceedings of the 8th International Conference on Frontiers for WWW Research and Development - APWeb 2006, Springer Berlin Heidelberg, LNCS.
