Research On Quality Prediction Of Complex Products Based On Improved LASSO-RF

With the continuous development of technologies such as intelligent manufacturing,Internet of Things,and big data,as well as the arrival of personalized,diversified,and customized consumption methods,the difficulty of quality control of complex products manufacturing process is greatly increased,which not only leads to the lower product qualification rate of complex product manufacturing enterprises,resulting in high inferior product production costs and quality characteristics testing costs,but also affect the timely delivery rate of products and the after-sales service costs.Therefore,the quality assurance problem of complex product manufacturing process has attracted more and more attention.How to effectively identify the Critical-to-Quality Characteristics(CTQs)of complex products manufacturing process and predict the quality of complex products has become a hot research topic in the field of quality management at home and abroad.For complex products,a single-piece small batch production method is generally adopted,the process and manufacturing process are complex,the raw materials and rework costs are high,and the quantity difference between qualified products and unqualified products is large.The quality characteristics data set of complex products has the characteristics of high dimension,small samples size,and data imbalance.This paper proposes a complex product quality assurance method based on quality prediction.The method avoids and reduces the quality loss and rework risk caused by unqualified product quality by establishing an improved Lasso-RF(Random Forest)product quality prediction model and predicting the quality of the product through manufacturing process data and product quality data.In view of the characteristics of complex products quality characteristics data set,in order to effectively identify the Critical-to-Quality Characteristics of the manufacturing process and reduce the complexity of quality prediction,this paper proposes the Bootstrap-K-split Lasso integrated feature selection algorithm(B-K-Lasso algorithm),which increases the stability and validity of feature selection.In order to reduce the impact of dataimbalance on the final prediction results,this paper uses the SMOTE algorithm to balance the data set,and then selects the RF integrated classification algorithm for product quality prediction.Finally,a complex producst quality binary classification prediction model based on B-K-Lasso RF was constructed,and simulation research and comparative analysis were carried out.Simulation research shows that the B-K-Lasso integrated feature selection algorithm can effectively identify the Critical-to-Quality Characteristics of complex products manufacturing process,reduce the dimension of the quality characteristics data set,and improve the stability of feature selection.When the value of the parameter K is appropriate,the classification performance of the subsequent RF integrated classification algorithm is better than that of the Lasso,K-split Lasso,and Bootstrap Lasso algorithms.For complex product quality prediction,especially for quality characteristics data sets with data imbalance characteristic,the quality prediction model based on B-K-Lasso RF is superior to the quality prediction model based on B-K-Lasso SVM,The RF integrated classification algorithm is more suitable for complex product quality prediction.