A random forest based machine learning approach for mild steel defect diagnosis

摘要

Industries today need to stay ahead in competition by servicing and satisfying customer's needs. Quality of the produced products to match as per customer demands is the key goal for a product manufacturing company. A product produced with variation in characteristics, than the anticipated are called as defect. In the mild steel coil manufacturing plants, large amount of data is generated with the help of many sensors deployed to measure different parameters which can be used for defect diagnosis of the coils produced. In case of mild steel coil, deviation of the final cooling temperature of the coil from desired temperature produces defective coils. The paper presents machine learning approach and the methodology for cooling temperature deviation defect diagnosis that consists of four phases namely data structuring, association identification, statistical derivation and classification. We also provide comparative results obtained with various data mining algorithms like decision trees, neural networks, SVM, ensemble techniques (boosting and random forest) in terms of performance parameters and prove that random forest outperforms rest of the techniques by achieving an accuracy of 95%.