Research On Key Algorithm Of Acoustic Fault Detection Of Chemical Equipment Based On CNN And RNN Network

In chemical production,equipment is usually subjected to harsh environments such as extreme temperature,pressure and corrosion,and is prone to various failures.These failures may lead to serious consequences such as equipment damage,production line downtime,product quality degradation,environmental pollution,and personal injury or death.Therefore,chemical companies must take measures to ensure the safety and reliability of equipment,including regular maintenance,equipment testing,failure analysis and safety training.This can effectively reduce the equipment failure rate and accident probability,and improve the safety and stability of the chemical production line.Fan is a general equipment widely used in all production links of chemical industry,but its complex working state and high speed rotation and other factors make it prone to various failures,so the fault detection research of chemical equipment in this paper takes fan as research object.In order to meet the needs of enterprises for fault detection,detection algorithms based on electrical,vibration and video technologies have been proposed.However,these algorithms are difficult to fully meet the requirements of practical detection.Acoustic sensors are able to measure objects in a non-contact manner,exhibiting high accuracy in measuring distance,position and velocity,etc.In addition,the acoustic sensor also has good anti-interference performance,can work normally in complex environments,and has a long service life and high reliability.From a practical standpoint,the acoustic sensor is small in size and light in weight,making them easy to integrate into systems and also very convenient to install and use.Therefore,based on acoustic signals,this paper studies a data-driven fault detection technology with strong adaptability,high real-time performance,good scalability and relatively easy to understand.It overcomes the traditional model-based and knowledgebased methods’ dependence on prior knowledge of experts or models defects.The research focus of this article is the application of deep learning technology in the field of chemical equipment fault detection.The main research contents are as follows:(1)Firstly,it is analyzed that the acoustic signal of the fan equipment has the characteristics of high frequency,noise interference,complex and changeable signal,high instability and load change.When performing acoustic fault detection of fan equipment,it is necessary to effectively process and analyze the acoustic signal to improve the accuracy and reliability of detection.Therefore,for the sound generation mechanism of the fan,Log-Fbank feature with good robustness in spectrogram was selected.This feature maps the time-domain and frequency-domain information into spatial and temporal information very well.(2)In view of the unbalanced characteristics of the data set,this paper performs data enhancement step by step from the two perspectives of acoustic signal and image.First of all,the original acoustic signal belongs to one-dimensional time series information,and the background noise is mostly Gaussian white noise,so the acoustic signal is transformed into the wavelet domain for threshold denoising and then inversely transformed to reconstruct the signal to complete the acoustic signal enhancement;secondly,for the spectrogram Log-Fbank,separately enhanced using histogram equalization and image pyramid techniques.(3)Based on the selected Log-Fbank features,this paper proposes a CNN-GRU fusion algorithm for wind fan fault detection.The algorithm takes into account the ability of CNN to process spatial pixels and the ability of GRU to process time series information,and can model the time domain and frequency domain information of acoustic signals well,so as to obtain a fan fault detection model with high confidence.(4)In order to further improve the accuracy of fault detection,use limited computing resources to quickly meet the computing needs of important parts,and reduce the requirements for hardware platform deployment of detection algorithms.This paper introduces a variety of attention mechanisms into the CNN-GRU algorithm model.Experimental results show that the CNN-GRU algorithm incorporating mixeddomain attention performs best in fault detection,while other attention mechanisms can also effectively improve detection results.