Research On Data-driven Product Quality Prediction Of Automotive Instrument Cluster

As an essential key auto accessory,the automotive instrument cluster intuitively reflects the real-time operation of the vehicle and the working status of the on-board systems.The quality of instrument products has a direct impact on vehicle performance and driving safety.At present,quality control of instrument products is mainly carried out through a number of function inspection links,and it is unable to intervene in advance of potential quality problems in the production process.Quality prediction enables field personnel to grasp the changing trend of product quality in advance and take the initiative in product quality control.Due to the continuous improvement of instrument product function and the increase of quality influencing factors in production process,the traditional quality influencing factor analysis method is difficult to obtain satisfactory prediction results.By establishing the relationship between process variables and target objects,data-driven method can effectively improve product quality control process.Therefore,this paper takes the production process of automotive instrument cluster as the research object.According to the production characteristics of multi-variety and smallbatch of instrument products,the data-driven automotive instrument cluster product quality prediction method is studied.The main work is as follows:Firstly,combining with the structure characteristics of automotive instrument cluster,the quality influencing factors of solder paste printing process and component assembly process in the production process of instrument products are analyzed.On this basis,a data-driven automotive instrument cluster product quality prediction framework is proposed,including instrument raw production data acquisition,screening of key quality influencing factors,and instrument product quality prediction modeling.Then,the sensitivity analysis of the quality influencing factors of automotive instrument cluster is carried out by feature engineering method,and the key quality influencing factors of instrument products in the production process are obtained.Firstly the raw production data is obtained through automated production equipment in the Surface Mounting Technology(SMT)and component assembly workshop.Then the raw production data is eliminated and normalized.The key quality influencing factors of instrument products are further analyzed and selected through expert experience,distance correlation coefficient and random forest method.Afterwards,according to the SMT process of PCB(Printed Circuit Board)assembly,which are the key functional parts of automotive instrument cluster,a solder paste printing volume prediction model based on Dense Convolutional Network(Dense Net)is proposed,providing technical support for ensuring the normal operation of various functions of instrument products.The Dense Net layers are designed to achieve the accurate prediction of key quality indicators of solder paste printing in SMT process,and the accuracy of the constructed prediction model is verified.On the basis of the above research,aiming at the imbalanced distribution of product quality data in the assembly process of automotive instrument cluster,an instrument product quality prediction method on imbalanced production data is proposed.The imbalance of quality data is alleviated by the improved Max Distance Synthetic Minority Over-sampling Technique(MDSMOTE).The Support Vector Machine(SVM)is used to establish the quality classification prediction model of instrument products,and the model hyper parameters are selected through the improved Particle Swarm Optimization(PSO)to improve the prediction performance.Finally the effectiveness and generalization are verified on different types and different batches of product quality data sets.