TL;DR
This paper tackles the challenge of evaluating the performance of ship block manufacturing processes (BMPs) by integrating process mining (PM) and data envelopment analysis (DEA). By analyzing actual work logs, the authors develop a systematic approach to identify underperforming BMPs and provide actionable insights for improvement, offering a more practical and accurate method for this complex industrial setting.
What Problem Does This Paper Address?
Shipbuilding is a complex process involving the assembly of hundreds of individual ship blocks. The efficient manufacturing of these blocks (Block Manufacturing Process - BMP) is crucial for overall shipbuilding productivity, cost reduction, and timely delivery. However, evaluating BMP performance is difficult due to several factors:
- Vast Number of Block Types: Ships are composed of numerous blocks, each with unique structures and manufacturing processes.
- Discrepancies Between Planned and Actual Work: Real-world manufacturing often deviates from planned schedules due to delays, urgent tasks, and resource issues, leading to significant time gaps (5-9 days) between planned and performed work.
- Heterogeneity of Processes: Even blocks of similar types can have variations in their manufacturing sequences and resource utilization.
Existing methods often focus on shipyard-level performance or general productivity, failing to provide granular insights into the specific BMPs. This paper aims to bridge this gap by offering a systematic and integrated approach to evaluate BMP performance based on actual operational data.
Methods
The proposed approach integrates two key methodologies:
- Process Mining (PM):
- Data Extraction: Log data is extracted from production information systems (e.g., BAMS). This data includes activity, time, identification, schedule, and material information.
- BMP Generation: PM is used to reconstruct the actual operational flow (sequence of activities) for each block from the extracted log data. This creates a "process model" for each BMP.
- Clustering: To enable fair performance comparisons, similar BMPs are grouped together. This is achieved by calculating similarity based on two vectors:
- Task Vector: Measures similarity in terms of the presence or absence of the same operations.
- Transition Vector: Measures similarity in terms of the sequential relationship of operations.
- The K-means clustering algorithm is then applied using these similarity measures to group homogeneous BMPs.
- Data Envelopment Analysis (DEA):
- Performance Metrics: Based on expert surveys and log data, several performance metrics are identified: total execution time, waiting time, gap between planned and actual working time, number of unit operations, and material amount.
- DEA Model: A modified DEA model with "target measures" (developed by Lim and Zhu, 2013) is employed. This model is particularly suited for handling metrics like the "gap between planned and actual working time," which can be positive or negative and has a desired target of zero.
- Input/Output Classification: Total execution time and waiting time are treated as DEA inputs (lower values preferred), while the number of operations and material amount are treated as DEA outputs (higher values preferred). The "gap" metric is handled as a special "target measure."
- Efficiency Score: The DEA model calculates an efficiency score for each BMP (Decision Making Unit - DMU). A score of 1 indicates a best-practice BMP, while a score less than 1 signifies an underperforming BMP.
Datasets
The study utilizes real log data from a Korean shipbuilding company, specifically from two projects exported from their Block Assembly Monitoring System (BAMS). This data represents actual work logs for block manufacturing. Eighty-six blocks were analyzed, and the data was used to generate BMPs, cluster them, and evaluate their performance. No new benchmark datasets were introduced; rather, the paper leverages existing industrial data.
Key Insights and Ideas
- Bridging the Gap with Integrated PM and DEA: The core innovation is the synergistic combination of PM to reconstruct actual processes from raw logs and DEA to evaluate their performance. This moves beyond theoretical models to data-driven, real-world performance assessment.
- The Importance of Process Similarity for Fair Evaluation: Recognizing that comparing dissimilar processes is misleading, the paper's clustering approach based on task and transition similarity is crucial for meaningful DEA.
- Handling the "Gap" Metric: The paper's adaptation of DEA to incorporate a "target measure" for the planned vs. actual working time gap is a significant contribution. This allows for a more nuanced evaluation than simply treating it as a standard input or output.
- Actionable Insights for Improvement: The DEA results not only identify underperforming blocks but also provide specific quantitative targets for improvement (e.g., reducing total execution time by X hours, waiting time by Y hours). This moves from diagnosis to prescription.
- Real-World Applicability: The case study demonstrates the practical utility of the proposed method in a real shipbuilding environment, with shipbuilding experts validating the findings and the usefulness of the provided guidelines.
Key Results
- Identification of Underperforming BMPs: The study successfully identified best-practice and underperforming BMPs within the analyzed dataset. For instance, in cluster C5, five blocks were identified as best-practice, while 14 were underperforming.
- Quantitative Improvement Targets: For underperforming blocks like 1XXX_110, specific targets were provided: reduce total execution time by 68 hours, waiting time by 40 hours, and the gap between planned and actual work by 7.1 days.
- Root Cause Analysis: The analysis highlighted that waiting time and total execution time were critical factors contributing to underperformance. For block 1XXX_110, issues like unnecessary transports to stockyards and extended waiting times were identified as key problems.
- Process Deviations Identified: The study showed that underperforming BMPs often deviate from their planned operational flows, sometimes due to worker discretion in choosing similar but not identical operations or workshops.
Limitations and Future Work
- Limitations:
- Process Structure-Based Classification: The paper acknowledges that BMPs were classified primarily based on process structure. For more reliable classification, a broader range of characteristics should be considered.
- Guideline Granularity: While guidelines for improvement are provided, they are considered a first step. More detailed, long-term, "best practices" strategies for stepwise performance improvement could be further developed.
- Future Work: The authors suggest that the provided guidelines can serve as a valuable starting point for developing additional strategies to enhance performance and productivity in block manufacturing. Further research could focus on refining the classification criteria and developing more comprehensive improvement strategies.
Summary updated