From Statistical Process Control to Smart Process Control : Developing an Explainable AI System for Manufacturing Quality Assurance
Pysyvä osoite
Kuvaus
Opinnäytetyö kokotekstinä PDF-muodossa.
In this thesis, a SPC dashboard that uses AI to enhance the quality of manufacturing was developed. The system incorporated the conventional SPC with ML anomaly detection, explainable AI, and an LLM interface. The four layered architecture was implemented successfully. Crimp Force Curves data set was used to validate the system.
Their findings indicated that the system is effective in detecting anomalies. The 38 violations that SPC identified had a Cpk of 0.061 which means that the process was critically out of control. ML identified 96 anomalies, and Isolation Forest was the best. The three models flagged nineteen samples, which are the most confident anomalies. The hybridization of SPC and ML was proved. SPC identifies clear differences, whereas ML detects subtle patterns controlling charts that are not detected by control charts.
The methods of explainability gave an insight into the anomalies. SHAP experienced a problem with dimension mismatch and thereby also used the d of Cohen as a default. Cohen d analysis indicated that all the features had a constant lower value in anomalies indicating a systematic issue. This was probably caused by a lack of force when crimping. Such an event proves the usefulness of explainability in detecting root causes.
Natural language explanations were offered by the LLM chat assistant. It has rightly cited the dashboard information and provided practical suggestions. This made the system operator-friendly and minimized the technical training. Technical precision and easy communication are a combination that is significant to adopt in practice.
The thesis makes three contributions. It shows the initial combined SPC-ML-XAI-LLM quality manufacturing platform. It offers an open-source software prototype of quality control education. It uses three machine learning models to compare on actual manufacturing data. These works contribute to the development of the sphere of smart process control.
The results substantiate that ML complements SPC with the ability to identify intricate trends. The unified system fills the gap of trust between the human operators and AI systems. The future of quality manufacturing is in human expertise and AI power, which necessitates explainable and accessible intelligent systems. The open source code can be used in the future research and education. The system can be reconfigured to other manufacturing processes. This has been the groundwork in making more developments of smart process control.