Research On Hydraulic Pump Assembly Quality Detection Based On Machine Learning

As the cost of labor increases,the demand for automation and intelligence in assembly lines is increasing.Especially in the production assembly lines of some micro-precision parts,it is difficult to meet the requirements of high strength and high precision by manual operation.In this paper an efficient and intelligent detection method for the assembly quality of hydraulic pump parts is proposed.It detects the mistakes in the assembly process of hydraulic pump blade parts by machine learning.In the hardware system chapter,according to traditional machine vision system,a suitable hardware system including the camera,light source,lens and host computer for the hydraulic pump parts is designed.The key parameters for each hardware in the detection system and the criteria for selecting each hardware are given.Considering all the cost and performance a reasonable detection system based on machine vision is designed.In the image segmentation chapter,the blade extraction algorithm from the hydraulic pump image is studied.Firstly,the commonly used image segmentation methods,such as threshold segmentation and edge segmentation,are introduced.Then,by analyzing the characteristics of assembly image of hydraulic pump,it is found that the image of hydraulic pump parts itself has rich geometric features,such as circle and rectangle.Using the feature that the base image is a circle,it is proposed to detect the image of the base and the blade by the Hough Circle Detection method.Then,the image is transformed into a polar coordinate system by coordinate transformation.Finally,the blade image is extracted from the base and blade image in the polar coordinate system by pixel projection.In the image feature chapter,the corner point feature and the Local binary patterns(LBP)feature in the image feature are first introduced.The characteristics of the blade image in different installation states and the application range of various features are studied.Then a new feature Gabor gray level co-occurrence matrix hybrid feature is proposed to identify the assembly quality of the blade and the effect is improved by 10% compared with other features.In the classifier chapter,the differences of classical pattern recognition classifier methods including Logistic Regression classifier and Support Vector Machines(SVM)classifier are compared.SVM classifier is used to detect the quality of the blade assembly considering the shortage of the hydraulic pump parts quality image samples.The detection accuracy of the assembly quality detection system of hydraulic pump reaches 0.916.Finally,we use online learning system and classifier updating strategy in the object tracking field for reference and design the support vector machine online learning method to make the classification system more intelligent.Our system performs better in terms of adapting to changes in the external environment and reducing equipment maintenance costs.