Research On Detection Technology Of MEMS Sensor Based On Machine Vision

With the rapid progress of micro-electronics,integrated circuits and the arrival of the Internet of things,Micro-electro Mechanical System(MEMS)is rapidly occupying the market share with its advantages of miniaturization,lightweight and stability,it became the mainstream sensor products in consumer electronics,automobile industry,chemical industry and medical field.However,due to the small size of MEMS components and their vulnerability to environmental impacts,higher requirements are put forward for MEMS component processing and detection technology.According to the research,most MEMS sensor manufacturers still mainly use artificial detection method in the detection process,which makes the detection process have the defects of low efficiency,high cost and low accuracy caused by human factors.Therefore,it is significative to develop and design an efficient detection technology of MEMS sensor.In this paper,by analyzing and researching object detection technology methods at home and abroad,a MEMS sensor visual detection technology based on machine vision is designed and developed to improve the efficiency of MEMS sensor production.The main content of this paper is as follows:(1)Visual inspection system hardware structure and software system design.According to the structure and physical characteristics of the detection target,the design and selection of components in the image acquisition device,such as light source,lighting mode,CCD industrial camera and lens,are completed.Using LED strip light source and low angle forward lighting as the solution of image acquisition lighting in this paper.It can eliminate the influence caused by metal surface reflection and complete the image acquisition of the workpiece under test.Modular design is carried out for the vision detection system studied in this paper,and the MEMS sensor chip size measurement and defect detection function are programmed(2)Application research of image processing algorithm.In this paper,the commonly used digital image processing algorithms such as image correction,enhancement,smoothing,feature extraction and other research,and for the median filter and Canny edge detection operator to improve the optimization algorithm,it has better processing precision through experiments.The final design proposes a set of image processing algorithm processing flow,realizes the accurate extraction of workpiece features and completes the preparatory work of MEMS sensor size measurement and defect detection.(3)Research on defect detection classification algorithm.In order to solve the problem of surface defect detection of MEMS components,this paper uses the defect feature matching based detection algorithm and the convolution neural network based defect detection algorithm to verify the comparison experiment.According to the experimental results,a defect detection and classification algorithm based on convolution neural network model is selected.A convolution neural network with eight layers structure is designed to solve the problem of difficult convergence of small sample image classification.By constructing the training sample database and completing the iterative training parameters,the defect detection and classification model meeting the detection requirements of accuracy and recall.(4)Visual inspection system experimental analysis and research.By using a batch of MEMS sensors for visual detection experiments,it is verified that the measurement accuracy of workpiece size,chip surface defect identification accuracy and measurement time of the visual detection system studied in this paper have met the requirements of visual detection system.At last,the error source is inferred according to the detection results,which provides the research direction for the subsequent system improvement.