Methodologies For Acoustic-based Diagnosis Of Rotating Machinery Under Industrial Environmental Condition

As an important transmission equipment,rotating machinery is widely used in mechanical systems of transportation,electric power and petroleum industrial fields for national economic construction and production development.In recent years,with the development of intelligent manufacturing,the real-time monitoring and fault diagnosis relying on the acquisition and representation of physical signal like acoustic,optics,electricity,vibration,mechanics and temperature is beneficial for manufacturing enterprises to promote production efficiency,reduce maintenance costs and improve economic benefits.However,due to the limitations of measurement,equipment structure and working condition,most of the existing contact-based fault diagnosis system are not available for each diagnosis task of rotating machinery in practical application.As a typical non-contact diagnosis technology,acoustic-based diagnosis(ABD)has been gradually explored for machinery fault detection due to its ability to overcome the limitation of contact-based diagnosis method through non-contact measurement by air-couple.It can supplement or partially replace the existing contact-based fault diagnosis system to make up for its shortcomings in some specific diagnosis tasks and application scenarios.Base on this consideration,a systematic and comprehensive deep learning-based ABD research for rotating machinery intelligent diagnosis under different working conditions in real-industrial scenarios is carried out.The main research content of this dissertation is summarized as follows.(1)the technical advantages and development potential of ABD method in intelligent fault diagnosis task is briefly introduced.The existing intelligent acoustic diagnosis technology is summarized,and the deficiencies in current works are analyzed,which provide a clear direction for our research content and technical route.Thus,the background concept of “ complex industrial environment” in our ABD study is expanded into three specific task including variable working condition,nonstationary background noise environment and their interactive coupling condition for hierarchical and progressive exploration.(2)The fault perception mechanism based on air-coupled acoustic signal is discussed in detail.A systematic illustration of the generation,propagation and perception acquisition of radiated noise from rotating machinery is implemented to lay a theoretical foundation for the subsequent ABD research.(3)Following the basic scientific research idea,the ABD study for rotating machinery under unstable working condition including variable load and speed,which involve the transformation in amplitude and phase of acoustic signal to cause amplitude modulation and frequency modulation phenomenon,is conducted in progressive manner by controlling interference factors.A novel ABD method based on multi-scale convolutional learning structure and attention mechanism is first proposed for rotating machinery fault diagnosis under different load conditions.The multi-scale convolutional learning structure is designed to automatically mine multiple scale features by different filter banks from raw acoustic signals.Subsequently,the novel attention mechanism,which is based on multi-scale convolutional learning structure,is established to adaptively allow multiscale network focus on relevant fault patterns information under different load conditions.In this way,the relationship between fault feature representation under different load conditions can be explored for promoting diagnosis knowledge transfer.Thus,the regularization ability of the ABD model can be improved to realize crossdomain diagnosis under different load conditions.On this basis,to perform ABD task under variable working condition for single-source to multiple-targets(SSMT)scenario,a novel hierarchical adversarial multiple-target domains adaptation(HAMTDA)learning framework is proposed in this dissertation.The hierarchical adversarial mechanism is constructed to improve the domain adaptation ability of ABD model against domain shift phenomenon in the interference of variable load and speed conditions by continuously exploring the domain invariant features through multistage domain adaptation learning.The experiment result in variable working conditions indicate that the proposed method has better robustness and generalization than other popular diagnosis method for rotating machinery cross-domain diagnosis under SSMT setting.(4)The strong and highly non-stationary background noise of real-industrial scenario is another challenging condition for ABD task.Therefore,the way to explore the noise suppression ABD method and anti-noise ABD method for rotating machinery are two primary mission in this study.Subsequently,a comprehensive acoustic diagnosis scheme based on noise suppression and anti-noise strategy is further developed to improve the diagnosis performance of ABD system in the interference of background noise.A novel noise suppression ABD method based on recursive denoising learning(RDL)is first proposed against noise interference.A new multi-stage attention mechanism is designed as fundament of RDL for adaptive tracking and estimating non-stationary industrial background noise and automatic suppressing noise.Based on the multi-stage attention mechanism(MSAM),a novel recursive learning strategy is introduced to further improve the performance of noise suppression by recursive tracking noise component and gradual denoising in coarse-to-fine manner.Meanwhile,a novel anti-noise ABD method based on recursive attention mechanism(RAM)is designed for rotating machinery.In proposed method,a recursive learning strategy is introduced to construct RAM by reusing the MSAM for multiple blocks to gradually estimate the noise interference probability within time-frequency(T-F)and adaptively simulated the corresponding interference in diagnosis model for enhancing anti-noise diagnosis ability.Finally,a novel two-stage ABD system,which performs denoising and anti-noise diagnosis in a step-by-step fashion,is proposed in this dissertation.In this method,the denoising sub-system can adaptively track and recursively suppress real-industrial background noise from collected multi-channel signal by designed recursive multi-head self-attention(MHSA)mechanism.Subsequently,a spatial multi-head self-attention(SMHSA)mechanism is explored to automatically estimate the noise interference probability within time-frequency(T-F)unit and interacts with diagnosis model in anti-noise diagnosis sub-system to further improve the diagnosis performance of ABD system.Based on the above research,a novel hierarchical adversarial ABD method based on knowledge distillation is proposed for rotating machinery diagnosis task under coupling interference between variable working condition and non-stationary noise environment.In this method,a hierarchical adversarial framework is designed to conduct multi-stage adversarial learning for obtaining domain invariant features.Then,an attention-based knowledge distillation mechanism is integrated into the framework to promote diagnosis knowledge transfer over target domains by knowledge extraction.Thus,the entire ABD model can be obtained to realize multi-target crossdomain diagnosis in complex coupled interference environments.Overall,several air-coupled acoustic-based intelligent methods are continuously explored for rotating machinery diagnosis task under different conditions following with the real-industrial scenarios.Their feasibilities and effectiveness are verified by specific simulation experiment.Thus,a systematic acoustic diagnostic technique scheme for gear of rotating machinery is formed in this dissertation,which provides reference for the subsequent theoretical and applied research.