PCB Board Drill Bit Defect Detection Based On Improved ELM Algorithm Research And Implementation

With the change and development of science and technology,the increasing demand for printed circuit boards in various industries,due to the complex production process of printed circuit boards,high production accuracy requirements.Drilling process is one of the extremely important links in the production process of printed circuit boards.As the drill bit is easy to wear or even collapse during the working process,which leads to product specifications not meeting the standards,how to quickly and effectively detect the failure of the drill bit or not has become a problem that must be paid attention to in the production process of printed circuit boards.To address the above issues,the main research work and contributions of this thesis are as follows:1)Using online weighted ELM to analyze the time-frequency domain characteristics of the sound signal of printed circuit board drilling for drill bit defect detection.By analyzing the time-frequency domain characteristics of the sound signals generated during drill bit drilling and using wavelet packet decomposition technology,the spectral features of the drill bit in each operating state are extracted and an ELM detection model based on sound feature volume identification is constructed.Based on the standard ELM,the online weighted ELM firstly introduces a weighting mechanism based on a cost-sensitive learning strategy,and gives each training sample a different misclassification cost weight in its cost function to alleviate the impact of imbalance of sample category labels on the cost calculation and network Second,the incremental learning method is used to fine-tune the model parameters in real time according to the dynamically changing sample data,so that the model can respond to the changes of sample data in time.The experiment proves that the improved online weighted ELM can effectively solve the above problems.2)Analyze the drilled hole diameter image by active online ELM to determine whether there are defects in the drill bit.For the aperture images that are easy to produce in the PCB drilling process,which cannot be classified by online ELM,an active online ELM is proposed.active online ELM is based on the standard ELM and introduces active learning strategy,firstly,the more valuable sample images are manually labeled and made into a training set,and the aperture edges are extracted by Canny edge detection and then trained iteratively to establish a detection model;secondly,in the detection process,important images that cannot be recognized by the model are again manually labeled and merged into the training set for iteration,thus continuously strengthening the model detection performance.The experiments prove that the improved active online ELM has high detection accuracy.3)To address the problem that the sound of each drill is intertwined in the drilling process of large-scale PCB production and the online weighted ELM cannot accurately determine the faulty drill,a hybrid fault detection method based on online weighted ELM for sound signal analysis and active online ELM for image analysis is proposed.The method firstly uses online weighted ELM to analyze the sound generated by multiple drills to determine whether there is a faulty drill in the drilling machine;secondly,it uses active online ELM to detect the image produced by the drilling machine with a faulty drill,so as to pinpoint the faulty drill.The experiment shows that the method can effectively detect the faulty drill.