Construction Of Digital Twin System For Fault Diagnosis Of Air Conditioner Based On Deep Learning

With the proposal of " deepening the implementation of the manufacturing power strategy”in the Outline of the 14th Five Year Plan,it is urgent to improve the level of informatization and intelligence in the manufacturing industry,and there is an urgent need for the deep integration of new generation information technology and manufacturing.As a key link in product quality control in the manufacturing industry,fault diagnosis in the production process also needs to be transformed from manual labor to intelligent detection and diagnosis,and the transformation includes two key points:intelligent diagnostic algorithms and intelligent control of diagnostic processes.Due to the powerful generalization ability and processing ability of massive heterogeneous data in deep learning,it has become a new method to solve the problems of low accuracy and efficiency in manual fault diagnosis of current production lines,and promote the intelligent transformation of fault diagnosis algorithms.The current fault diagnosis process control relies excessively on human factors,and the intelligent control methods for human,machine,material,and material in the production process are insufficient.The physical space and control information space lack real-time information interaction,making it impossible to form an intelligent and interconnected software and hardware real-time interaction control system.As an emerging intelligent technology for virtual real mapping and real-time interaction,digital twin,combined with end-to-end cloud collaboration technology,constructs an intelligent real-time digital twin system,which is a new approach to improve the intelligence of fault diagnosis process control.Based on the above background,this article takes the fault diagnosis of air conditioner internal units as an example,collects noise data of air conditioner internal units and cleans them,constructs a fault diagnosis digital twin system based on deep learning,and conducts the following innovative research:(1)In response to the problems of low algorithm accuracy and efficiency in traditional fault diagnosis methods,this article constructs a deep learning algorithm based fault diagnosis method for air conditioner internal units.It proposes a comprehensive solution from data collection,data cleaning,model training to fault diagnosis,which improves the efficiency of fault diagnosis.In addition,in order to promote research on fault diagnosis in industrial scenarios,a noise dataset for air conditioning internal units has been developed and made public.(2)To improve the effectiveness of sound feature extraction,in the process of cleaning sound data and extracting features,this paper proposes an improved time-frequency feature extraction method SG-Gram based on psychoacoustics.By introducing psychoacoustic spectrum to improve the time-frequency feature extraction method,the ability to extract sensitive sound features is enhanced.To improve the accuracy of deep learning fault diagnosis algorithms,this paper proposes improved methods of deep neural networks for sound recognition,ARResNet and ARDenseNet.After experimental verification,the proposed method improves the diagnostic accuracy to 96.97%.(3)In response to the problem of imbalanced training data caused by low probability of occurrence of some fault types,this article uses the Quasi-periodic Parallel Wave Generative Adversarial Network QPPWG to generate and expand low probability noise sample data,alleviating the problem of low probability sample fault diagnosis accuracy caused by data imbalance.After data expansion,the accuracy is improved to 86.66%.(4)To solve the problem of insufficient control measures for manual fault diagnosis processes,this paper constructs a set of digital twin systems for fault diagnosis of air conditioners based on deep learning,carries the deep learning algorithm proposed in this paper,realizes the full cycle full process virtual reality mapping and visual display of fault diagnosis,and provides an example of intelligent control for the Industrial Internet of Things scenario.To sum up,this paper conducts research on fault diagnosis of air conditioners in the IIOT scenario,builds a digital twin system for fault diagnosis based on deep learning,proposes improved sound feature extraction methods and deep learning algorithms ARResNet and ARDenseNet,and expands low probability sample mechanical sound data through QPPWG,which alleviates the problems of low probability sample learning difficulties and low fault diagnosis accuracy caused by data imbalance.The digital twin system in this paper provides an example for intelligent management and control in the IIOT scenario.