A Hybrid Data Reduction and Knowledge Extraction Algorithm for Quality Prediction

Background: With the explosive growth of the manufacturing data, the manufacturing enterprises paid more and more attention to dealing with the manufacturing big data. The manufacturing big data also can be summarized as "5Vs”, volume, variety, velocity, veracity and value. Recently, the researchers are focused on proposing better knowledge discovery algorithms to handling the manufacturing big data. Objective: The high dimensional data can be reduced from two directions. The one was the dimension reduction. It makes the data set simple and overcome the problem of curse dimensionality. This method reduced the data set form the data width. Methods: We proposed a hybrid data reduction and knowledge extraction algorithm (HDRKE) for quality prediction. There are 5 steps in the algorithm: Step 1: Data preprocessing; Step 2: Dimension reduction; Step 3: Extract SVs by SVM; Step 4: Extract rules from the subset; Step 5: Prediction by the rules extracted in step 3. Results: The presented HDRKE method reduced the data scales from the data dimensions and the data attributions. Then, the prediction method was used on the subset of reduced data. At last, the HDRKE method was applied to a enterprise sample, the validation of the method can be validated on the enterprise sample. Conclusion: Quality prediction and control was an important procedure in manufacturing. The HDRKE algorithm was a novel method based on the attribution reduction and dimensionality reduce. The data set simplified from double direction made the data set easily to calculate. The HDRKE method also proposed a new thought of decision rules extracting on the low-embeddings. The HDRKE method also applied to a manufacturing instance and proved its validity.