Series Production Quality Control With ML Root Cause Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Quality control in mass-produced products becomes increasingly difficult as the number of manufacturing steps increases, with final inspections being time-consuming and root cause analysis labor-intensive, and existing methods struggle to efficiently identify the underlying causes of quality issues.
Innovation Solution
A method involving the recording of manufacturing process parameters, using a machine learning model to predict quality, and an explainer to analyze deviations, allowing for rapid identification of root causes by mapping parameters to quality assessments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If quality control is performed after each manufacturing step, then product quality can be monitored continuously, but the inspection process becomes too time-consuming for mass production
Solution Approach 1:
The system performs preliminary quality assessment by recording manufacturing parameters and using a trained machine learning model to predict quality outcomes before final inspection. This allows most products to be cleared quickly based on parameter analysis, with only suspicious cases requiring full physical inspection, thereby maintaining quality monitoring while preserving production speed
2Loss of information
If root cause analysis is performed manually after final inspection, then quality problems can be investigated, but the process is labor-intensive and time-consuming
Solution Approach 1:
The system replaces manual root cause analysis with an automated machine learning model that has been trained on historical manufacturing data. When a quality problem is detected, the model automatically analyzes the recorded manufacturing parameters to identify the root cause, substituting labor-intensive manual investigation with automated computational analysis that is both faster and more consistent
3Measurement precision
If final production inspection is performed, then product quality can be assessed, but the specific issue discovered doesn't always provide a useful clue to its underlying cause
Solution Approach 1:
The system records all manufacturing parameters throughout the production process as preliminary data. When a quality issue is detected during final inspection, this pre-recorded parameter data is already available for analysis, providing clues about the underlying cause that would otherwise be lost. The machine learning model then processes this information to identify root causes
Solution Approach 2:
The system establishes a feedback loop where final inspection results are fed back into the machine learning model along with the recorded manufacturing parameters. This feedback enables the model to learn from actual quality outcomes and improve its ability to predict and diagnose quality issues, ensuring that information from final inspection is fully utilized rather than lost
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method (100) for the quality control of a product (1) produced in series, comprising the steps of: ◦ during the production of the product (1), a plurality of parameters (11) which characterize the production process is recorded (110); ◦ the product (1) is subjected (120) to a control, said control involving at least one physical observation of the product (1), and/or at least one physical functional test of the product (1), and involving a comparison of the result of the observation, or the result of the functional test, with a predefined reference, by means of which comparison a quality assessment (12) of the product (1) is determined; ◦ in response to the satisfaction of a predefined criterion (130) by said quality assessment (12), the parameters (11) recorded during the production of the product (1) are fed to a trained machine-learning model (2) and are mapped (140) to a quality prediction (13) for the product (1) by said trained machine-learning model (2); ◦ it is checked (150) whether the quality prediction (13) is consistent with the quality assessment (12); ° if this is the case, the recorded parameters (11) are fed (160) to an explainer (21) for the machine-learning model (2), which explainer (21) assigns to individual recorded parameters (11), and/or to combinations of the recorded parameters (11), quantitative contributions (14) to the quality prediction (13); ◦ on the basis of the quantitative contributions (14), a probable cause (15) of the quality assessment (12) obtained in the control (120) is evaluated (170).