An Experimental Diagnostic Procedure to Identify the Source of Defects in Multi-Stage and Multi-Component Production Processes

Many production processes consist of successive steps in which things can go wrong without notice because the problem is only detectable in the final product. For instance in steel manufacturing, the coils undergo melting, hot rolling, annealing and pickling, and defects in one of these stages only become visible after the final process. In other production processes, an output issue may only be detected during final testing after the different parts have been assembled. In all these cases, it is hard to determine which part of the production process is responsible for an unusually high defect rate. We describe a simple procedure based on cluster detection to identify the problematic step if the following conditions are satisfied: the production of defects tends to occur clustered in time and it is feasible to (partially) reorder the part or batch processing sequence in each stage of the production process. Even if reordering is not required for the production, the diagnostic information that can be obtained can well outweigh the potential extra costs involved.