| Since the 21st century,it is becoming a new trend to develop the modern medical equipment with aritifical intelligence,large scale integrated and modular circuit.Manufacturers no longer provide technical data such as circuit diagram and maintenance code intentionally so that they can obtain monopoly profits of after-sales maintenance.Note that the hospital maintenance engineers only have the basic skills on instrument maintenance.Since the theory and technology of medical equipment maintenance have not been developed accordingly,the hospital engineers are not well qualified to fix those instruments.The reasons why the engineers cannot fix machines can be summarized as follows:1.No technical data.Without complete technical data or systematic training provided by the manufacturers,the maintenance engineers cannot repair the medical equipment effectively due to the outdated experience and traditional tools.2.Only board card maintenance.When medical equipment is broken,the maintenance engineers use the theory of equipment composition,the function of circuit board and the tips of diagnosis system to repair the equipment.However,they can only locate the faulted circuit board,no the chip level.Then,they will contact the manufacturers to replace the whole piece of circuit board directly.There is no doubt that the maintenance cost increses.3.Maintaining the chip level is more critical.The maintenance engineers should check all components and electrical signals of the circuit board in case of any failure.This method not only reduces the maintenance efficiency but also demands high technical ability of the maintenance engineers.With the improper operation,it is easy to cause fault range expansion,which causes certain personal injury.4.Outdated maintenance tools.At present,the hospital engineers still use traditional maintenance tools such as multi-meter and oscilloscope to measure one feature of electrical signals at each point on the circuit board one by one.This operation is unable to view,analyze and process multiple features of electrical signals at the same time,and it is inefficient and not conductive to do troubleshooting by maintenance engineers.As a result,it is difficult to satisfy the maintenance requirements.Compared with the traditional fault diagnosis method mentioned above,these methods such as multivariate statistical analysis,signal analysis,artificial neural network,support vector machine and other shallow machine learning methods are widely used in equipment fault detection,but the extraction and selection of the fault features is mainly done by human experience.They have limited ability to express data and learn the internal characteristics of complex signals.In view of the above problems,this paper takes the monitor switch ing power supply,which is widely used in the market and has high failure rate as the research task,and study the fault diagnosis method based on convolutional neural network in deep learning.The work can be summarized as follows:1.This paper analyzed the basic composition and principle of switching power supply,as well as investigating and analyzing the common failure types and causes.We simulated the T8 monitor switching power supply three kinds of fault artificially,and then carried on the module division combining with the fault type and function testing principle.A total of16 test points were selected to acquire data.2.This paper built the multi-channel data acquisition system of "LabVIEW+ data collection card" based on LabVIEW platform,and then system hardware structure and software design and development were introduced.Under the condition of the whole operation of the monitor,the electrical signals corresponding to the switch ing power failure and normal state were collected to complete the data collection.3.The basic structure of convolutional neural network had been improved by introducing batch normalization.Therefore,the fault diagnosis model was established.For the deficiencies in the model training process,this paper optimized loss function by adding the regular term and optimized training process by stochastic gradient descent by mini-batch.4.The collected data of T8 monitor switching power supply in different state was divided into multiple samples.Among them,20000 samples were as training set,4000 samples were as a validation set and 4000 samples were as test set.The effects of different training parameters on the diagnostic results of the model were studied,as well as the visual analysis and multi-classification performance evaluation of the model.Finally,the model was validated robustly by adding noise and losing part of data.The research results had showed that the fault diagnosis model based on convolutional neural network could effectively identify the fault of the switching power supply,the average diagnostic accuracy of the model after multiple experiments was 99.7 9%.After adding noise,the diagnosis almost unchanged,the accuracy also could reach 99.63%,and a perfect result gained such as the precision rate and recall rate,this indicated that the robustness of model was stable.Therefore,the study of medical equipment fault diagnosis method based on convolutional neural network can improve the efficiency and accuracy of medical equipment fault diagnosis,and it is expected to solve the problem of difficult and expensive maintenance of medical equipment without drawing and other technical data,which is of great significance to guarantee the safe and effective operation of medical equipment. |
