Research On Intelligent Fault Diagnosis For Main Control Board Of Soft Tissue Therapy Instrument Based On Long Short-term Memory Network

As an important part of modern medicine,medical equipment is a necessary condition and important support for clinical medical treatment,teaching and scientific research.With the progress of science and technology and the rise of public demand for medica l treatment,the type and quantity of medical equipment are growing rapidly.It accounts for 55%-75%of the total fixed assets of the hospital,and continues to increase at an annual growth rate of15%-25%.Once a failure occurs,it will affect the efficiency of diagnosis and treatment,and seriously threaten the safety of operators and patients.Therefore,how to repair the fault timely and efficiently to ensure the safe and reliable operation of all kinds of medical equipment is particularly important.The traditional fault diagnosis and maintenance of medical equipment is usually based on the preliminary analysis of the fault phenomenon,combined with the known circuit drawings,using multimeter,oscilloscope,on-line tester.By measuring the electrical parameters of each test point,the circuit board fault is compressed and located with the help of existing maintenance experience and logic.In view of the fault diagnosis and maintenance of modern medical equipment,the traditional methods are faced with severe challenges:(1)The circuit of modern medical equipment has high integration and complex design.The design is generally modular,integrated,novel structure and complex layout,so it is difficult to judge the system structure and signal flow direction by using traditional maintenance experience,and it is difficult to carry out fault detection and diagnosis.(2)Manufacturers no longer provide drawings and other technical data.Manufacturers attach great importance to intellectual property rights,so as to provide only simple equipment system diagrams,general maintenance tools and training.Lack of relevant technical support,hospital maintenance engineer can not repair the fault equipment.(3)Manufacturers focus on the replacement of circuit board modules,the maintenance cost remains high.In order to obtain high maintenance monopoly profits,medical equipment manufacturers do not provide maintenance at the level of circuit board components.The technological monopoly of manufacturers in this area has further aggravated the difficult situation of medical equipment maintenance.In order to solve the above problems faced by traditional medical equipment fault diagnosis,inspired by the rapid development of artificial intelli gence and machine learning and extensive research and application in other fields.This thesis intended to select the main control board of YDB-III soft tissue therapy instrument as the research object which is widely used in medical institutions.Focusing on how to make full use of the potential timing and periodic characteristics of the electrical signal data of the circuit board port,as well as the supplementary information contained in the fault symptom phenomenon,the two were fused and processed.The Long Short-Term Memory(LSTM)network was selected to build a fault diagnosis model to study the method of intelligent fault diagnosis of medical equipment without drawings.The main work of this thesis was as follows:(1)Design of data acquisition system.Aiming at the research object of this thesis,firstly,under the condition that the technical data such as circuit drawings were unknown,the function of the external port was analyzed and tested,the interface board was designed,and the acquisition software was written.Under the same peripheral working conditions,the normal working state and six common fault states were selected,the symptoms of 7states were collected and recorded in detail.Based on the Lab VIEW program development environment,the NI data acquisition card was used to collect electrical signal data from 45 port channels of 30 main control boards at a frequency of 3000 Hz.A total of 210 groups of data were collected for model training and fault diagnosis.(2)Establishment of intelligent fault diagnosis model.With the LSTM as the core,Dropout and Early Stopping regularization terms were used to prevent the model from overfitting,and the gradient descent method with Nesterov accelerated gradient was used to optimize the model,and an intelligent fault diagnosis model was designed.It was verified by establishing four fault diagnosis sample feature sets: electrical signal feature set,fault symptom phenomenon feature set,electric signal and symptom phenomenon fused feature set,fused and screened feature set.In the experiment,every 300 data points were taken as a sample,each feature set was divided according to the proportion of 4:1:1,and the sample numbers of training set,verification set and test set were 1400,350,350 respectively.(3)Experimental verification and result analysis.The experimental verification of fault diagnosis was carried out based on 4 sample feature sets,specifically,the data of 7categories(1 normal state,6 fault states)were diagnosed and classified.By analyzing the accuracy,supplemented by confusion matrix,precision,recall,F1 value,AUC value,to judge the model diagnosis results and the influence of different feature sets on the results.In addition,compared with other common fault intelligent diagnosis methods,such as BP Neural Network(BPNN),Recurrent Neural Network(RNN),Convolution Neural Network(CNN),to compare the classification and identification of different network.The experimental results showed that based on the LSTM network model,using the fused and screened feature,the average accuracy was 97.09%,which was 8.41% higher than 88.68% of the electrical signal feature alone,24.98% higher than the 72.11% of the symptom feature alone,and 5.94% higher than the 91.15% of the fused feature.At the same time,the precision,recall and other evaluation indicators have been improved accordingly.Vertical comparison of different algorithms,in the case of using fus ed and screened feature,the accuracy of the diagnosis model designed in this thesis was 12.57%,4.71% and 0.57%higher than that of BPNN,RNN and CNN,respectively.To sum up,the intelligent fault diagnosis method based on long-term memory network and using the fused and screened feature of electrical signals and symptoms has higher accuracy.It can provide research ideas and technical support for further solving the fault diagnosis of medical equipment circuit board which is lack of technical data.