An air conditioner bearing fault diagnosis method and system based on acoustic emission signals

Abstract

The application provides an air conditioner bearing fault diagnosis method and system based on an acoustic emission signal, and the air conditioner bearing fault diagnosis method comprises the following steps: collecting original acoustic emission signals in a bearing operation process in real time; decomposing the original acoustic emission signals by using a CEEMDAN algorithm to obtain multiple IMF components; constructing a fault characteristic factor based on the fusion of kurtosis and a Pearson correlation coefficient, and screening out key IMF components; calculating the permutation entropy of the key IMF components, and constructing a bearing state feature vector after normalization; optimizing a support vector machine classification model by using a DSSA algorithm; inputting the bearing state feature vector into the optimized support vector machine classification model, and outputting an air conditioner bearing fault type recognition result. The application solves the technical problem that the existing diagnosis algorithm has limited feature extraction capability for complex and non-stationary signals and has a high misjudgment rate.

Description

Technical Field

[0001] This invention relates to the field of mechanical fault diagnosis technology, and more specifically, to a method and system for diagnosing air conditioning bearing faults based on acoustic emission signals. Background Technology

[0002] Air conditioning equipment has become an indispensable temperature control device in modern buildings, rail transit, industrial production, and household life. Its long-term continuous operation reliability directly affects the user experience, energy consumption level, and equipment life. As core rotating support components, the bearings of air conditioning compressors, indoor fans, and outdoor fans operate under harsh conditions such as high speed, alternating load, enclosed cavity, poor heat dissipation, and complex structural transmission paths. They are prone to early minor faults such as inner ring peeling, outer ring defects, rolling element wear, micro-crack propagation, and lubrication failure. If bearing faults are not detected in time, they can easily cause increased vibration, abnormal noise, and decreased efficiency. In severe cases, it can lead to compressor seizure or damage to the entire machine.

[0003] With the development of acoustic emission detection technology, acoustic emission signals, due to their high frequency, transient nature, and high sensitivity to local damage and impact, have been gradually used for early bearing fault detection. However, under actual air conditioning operating conditions, acoustic emission signals exhibit characteristics such as non-stationarity, strong noise, multi-scale coupling, weak fault characteristics, and complex background interference, resulting in a series of technical bottlenecks that are difficult to overcome in existing technologies.

[0004] However, the relevant technologies have at least one of the following problems: existing diagnostic algorithms have limited ability to extract features from complex, non-stationary signals, resulting in a high misjudgment rate. Summary of the Invention

[0005] This invention addresses the technical problem that existing diagnostic algorithms have limited feature extraction capabilities for complex, non-stationary signals, resulting in a high misjudgment rate.

[0006] To address the aforementioned problems, this invention provides a fault diagnosis method for air conditioner bearings based on acoustic emission signals. The method includes: real-time acquisition of raw acoustic emission signals during bearing operation; decomposition of the raw acoustic emission signals into multiple IMF components using the CEEMDAN algorithm; construction of fault characterization factors based on kurtosis and Pearson correlation coefficient, and selection of key IMF components; calculation of permutation entropy for the key IMF components, normalization, and construction of a bearing state feature vector; optimization of the support vector machine classification model using the DSSA algorithm; inputting the bearing state feature vector into the optimized support vector machine classification model, and outputting the air conditioner bearing fault type identification result.

[0007] Compared with existing technologies, the technical effects achieved by this solution are as follows: This application fully utilizes the high sensitivity of acoustic emission signals to early minor damage in bearings. By introducing a processing method combining CEEMDAN multi-scale decomposition and adaptive feature selection, noise interference is effectively suppressed and fault impact characteristics are highlighted, significantly improving the detection sensitivity of early faults in air conditioning bearings. Simultaneously, features constructed based on permutation entropy can stably characterize the nonlinear and complex features of acoustic emission signals, improving the reliability of feature extraction. Furthermore, the DSSA algorithm is used to globally optimize the support vector machine parameters, avoiding performance fluctuations caused by the reliance on empirical parameter settings in traditional classification models, thereby significantly improving the accuracy and generalization ability of fault identification. The overall method is applicable to complex operating conditions such as air conditioning compressors and fans, providing effective technical support for predictive maintenance and intelligent operation and maintenance of air conditioning equipment.

[0008] In one embodiment of the present invention, the DSSA algorithm includes: introducing a population diversity preservation strategy based on Euclidean distance to perform population diversity initialization; introducing a Gaussian perturbation strategy to update the discoverer position by adaptively adjusting the perturbation amplitude; and introducing a fitness difference guidance strategy based on global optimum to update the joiner position.

[0009] Compared with existing technologies, the technical effects achieved by this solution are as follows: Through a three-layer progressive structure of diverse initialization—discoverer global search—joiner local tracking, a strongly correlated and highly collaborative optimization mechanism is formed, comprehensively improving the overall optimization performance of the algorithm; the discoverer is responsible for searching the optimal region over a large area, while the joiner is responsible for finely converging the optimal value, with clear division of labor and efficient collaboration, significantly improving the speed and stability of SVM parameter optimization; the three-layer improvement strategy is mutually supportive and inseparable, making the DSSA algorithm particularly suitable for high-dimensional, nonlinear, multi-peak, and noisy fault diagnosis optimization scenarios, significantly improving the final classification accuracy of the model.

[0010] In one embodiment of the present invention, a Gaussian perturbation strategy is introduced to update the discoverer's position. This involves adaptively adjusting the perturbation amplitude, including: introducing a Gaussian distributed perturbation term to adaptively adjust the discoverer's position. The perturbation amplitude is dynamically adjusted with the number of iterations. ;in, The solution position for individual i after the update. Let i be the current solution position of individual i. This represents a standard Gaussian distribution, with a mean of 0 and a standard deviation of . The disturbance; This is the current globally optimal solution. This is the worst solution at present; The perturbation magnitude for the current iteration. The initial perturbation magnitude is given by `iter`, where `iter` is the current iteration number. This represents the maximum number of iterations.

[0011] Compared with existing technologies, the technical effects achieved by this solution are as follows: the introduction of Gaussian distributed random perturbation significantly improves the diversity of search directions for discoverers, expands the effective search range, and enhances the ability to escape local optima; the perturbation amplitude adaptively decays with the number of iterations: the amplitude is large in the early stage to ensure full global exploration; the amplitude is small in the later stage to ensure accurate local convergence, achieving a dynamic optimal balance between exploration and development.

[0012] In one embodiment of the present invention, a population diversity preservation strategy based on Euclidean distance is introduced to perform population diversity initialization, including: calculating the Euclidean distance between any two individuals in the sparrow population; and removing individuals in the sparrow population whose Euclidean distance is less than a set threshold to obtain a uniform population.

[0013] Compared with existing technologies, the technical effects achieved by this solution are as follows: This application uses a population diversity preservation strategy based on Euclidean distance to construct a uniformly distributed initial sparrow population, enabling the population to have sufficient global search potential and providing diverse initial directions for discoverer location updates; the uniform population provides a basic solution space distribution for subsequent discoverer searches and joiner tracking, achieving uniform initial population distribution, maximizing solution space coverage, and improving global search potential from the source; redundant and similar individuals are eliminated, high-diversity and high-quality initial solutions are retained, ineffective iterations are avoided, and the overall operating efficiency of the algorithm is significantly improved.

[0014] In one embodiment of the present invention, a fitness difference-guided strategy based on global optimum is introduced to update the joiner's position, including: calculating the fitness difference between the current individual and the global optimum: Dynamically adjust the search step size by combining fitness differences and the global optimal solution: Update the joiner positions with the goal of achieving the global optimal solution: .in, For individuals Differences in fitness For individuals fitness The fitness of the global optimal solution; Let j be the search step size. This is the step size adjustment factor. These are uniformly distributed random numbers; The solution position for individual j after the update. For individuals The current solution position, This represents the location of the globally optimal solution.

[0015] Compared with existing technologies, the technical effects achieved by this solution are as follows: The participant position update uses fitness differences as the basis for dynamic step size adjustment, directly inherits the global search direction provided by the discoverer, and takes the global optimal solution as the convergence target, achieving linkage control between the search direction and the search step size; Furthermore, it forms a strong correlation and collaborative mechanism with the discoverer position update: the discoverer provides the optimal direction, and the participant tracks quickly and accurately, significantly improving the overall algorithm performance; By strongly linking and binding the four key elements—fitness differences, global optimal solution, search step size, and participant position—it ensures that the participant always converges efficiently along the optimal direction determined by the discoverer.

[0016] In one embodiment of the present invention, the optimization of the support vector machine classification model using the DSSA algorithm further includes: using the SVM classification accuracy as the fitness function, iteratively optimizing the optimal penalty parameters and kernel function parameters through the DSSA algorithm; outputting the optimal combination of SVM parameters with the goal of maximizing classification accuracy; and optimizing the support vector machine classification model through the optimal combination of SVM parameters.

[0017] Compared with existing technologies, the technical effects achieved by this solution are as follows: Maximizing classification accuracy is the direct optimization objective, ensuring a high degree of consistency between the parameter optimization process and the fault diagnosis task, thus guaranteeing that the optimization results directly improve diagnostic accuracy; through the global optimization capability of DSSA, fully automatic intelligent optimization of SVM parameters is achieved without manual parameter tuning, effectively avoiding local optima and obtaining a better parameter combination than traditional methods, significantly improving the model's generalization ability and robustness; simultaneously, it supports online real-time updates of model parameters, adapting to monitoring scenarios of air conditioning equipment operating under varying conditions and long cycles, ensuring continuous high-precision diagnosis.

[0018] In one embodiment of the present invention, the CEEMDAN algorithm is used to decompose the original acoustic emission signal into multiple IMF components, including: adding adaptive white noise to the original acoustic emission signal to construct an auxiliary signal; performing multiple EMD decompositions on the auxiliary signal and taking the average value to obtain the first-order IMF component; calculating the first-order residual component based on the first-order IMF component, and continuing to add white noise for iterative decomposition to obtain each-order IMF component and each-order residual component, until the residual component is a monotonically indivisible signal; and representing the original acoustic emission signal as the sum of each-order IMF component and the final residual component.

[0019] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: CEEMDAN fundamentally suppresses mode aliasing by adding adaptive white noise step by step and averaging multiple times, ensuring that each IMF component is independent and clear and the fault information is complete; at the same time, it can retain weak early fault impact information, so that the transient characteristics generated by minor damage are not submerged by noise, and significantly improve the early fault detection capability.

[0020] In one embodiment of the present invention, a fault characterization factor is constructed based on the fusion of kurtosis and Pearson correlation coefficient, and key IMF components are selected. This includes: calculating the kurtosis value of each IMF component; calculating the Pearson correlation coefficient between each IMF component and the original acoustic emission signal; normalizing the kurtosis value and the Pearson correlation coefficient respectively and then weighting and fusing them to construct the fault characterization factor; and selecting the top m IMF components as key IMF components according to the fault characterization factors sorted from largest to smallest.

[0021] Compared with existing technologies, the technical effects achieved by this solution are as follows: by integrating the impact sensitivity of kurtosis with the global correlation of Pearson correlation coefficient, complementary advantages are achieved, significantly improving the accuracy of fault component identification and anti-interference ability; and by constructing adaptive fault characterization factors, key components containing the strongest fault information can be automatically screened without manual intervention, greatly improving intelligence and adaptability.

[0022] In one embodiment of the present invention, the air conditioning bearing failure types include: normal state, inner ring failure, outer ring failure, and rolling element failure.

[0023] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: it enables accurate diagnosis in four categories, clearly distinguishing between normal, inner ring fault, outer ring fault, and rolling element fault, with clear fault location and classification results that can directly guide maintenance; it provides accurate data support for equipment condition assessment, remaining life prediction, and maintenance strategy formulation, significantly reducing operation and maintenance costs, reducing unplanned downtime, and extending equipment service life.

[0024] On the other hand, embodiments of the present invention also provide an air conditioning bearing fault diagnosis system based on acoustic emission signals. The air conditioning bearing fault diagnosis system employs the air conditioning bearing fault diagnosis method based on acoustic emission signals as described in any embodiment of the present invention. The air conditioning bearing fault diagnosis system includes: an acquisition module for real-time acquisition of raw acoustic emission signals during bearing operation; a decomposition module for decomposing the raw acoustic emission signals into multiple IMF components using the CEEMDAN algorithm; a screening module for constructing fault characterization factors based on the fusion of kurtosis and Pearson correlation coefficient, and screening out key IMF components; a permutation entropy feature module for calculating permutation entropy for the key IMF components, normalizing it, and constructing a bearing state feature vector; an optimization module for optimizing the support vector machine classification model using the DSSA algorithm; and an identification module for inputting the bearing state feature vector into the optimized support vector machine classification model and outputting the air conditioning bearing fault type identification result.

[0025] Compared with the prior art, the technical effects achieved by adopting this technical solution are as follows: The air conditioner bearing fault diagnosis system based on acoustic emission signals in this embodiment is used to implement the air conditioner bearing fault diagnosis method based on acoustic emission signals as in any embodiment of the present invention. Therefore, it has all the beneficial effects of the air conditioner bearing fault diagnosis method based on acoustic emission signals as in any embodiment of the present invention, which will not be repeated here.

[0026] By adopting the technical solution of the present invention, the following technical effects can be achieved: (1) By fully utilizing the characteristic that acoustic emission signals are highly sensitive to early minor damage in bearings, and by introducing a processing method that combines CEEMDAN multi-scale decomposition with adaptive feature screening, noise interference is effectively suppressed and fault impact characteristics are highlighted, thus significantly improving the detection sensitivity of early faults in air conditioning bearings. (2) Features constructed based on permutation entropy can stably characterize the nonlinear and complex features of acoustic emission signals, thereby improving the reliability of feature extraction; (3) The DSSA algorithm is used to optimize the parameters of the support vector machine globally, avoiding the performance fluctuation problem caused by the dependence of traditional classification model parameters on empirical settings, thereby significantly improving the accuracy and generalization ability of fault identification. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 A flowchart of an air conditioner bearing fault diagnosis method based on acoustic emission signals provided in Embodiment 1 of the present invention; Figure 2 A detailed flowchart of an air conditioner bearing fault diagnosis method based on acoustic emission signals provided in Embodiment 1 of the present invention; Figure 3 for Figure 2 Detailed flowchart of the DSSA algorithm in China; Figure 4 This is a schematic block diagram of an air conditioner bearing fault diagnosis system based on acoustic emission signals, provided in Embodiment 2 of the present invention.

[0028] Explanation of reference numerals in the attached figures: 100-Air Conditioner Bearing Fault Diagnosis System; 101-Acquisition Module; 102-Decomposition Module; 103-Filtering Module; 104-Permutation Entropy Feature Module; 105-Optimization Module; 106-Identification Module. Detailed Implementation

[0029] To make the above-mentioned objectives, features, and advantages of the present invention more apparent and understandable, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1 See Figure 1 This is a flowchart of a method for diagnosing air conditioning bearing faults based on acoustic emission signals provided in the first embodiment of the present invention, combined with... Figure 2 Methods for diagnosing air conditioner bearing faults include: S100: Real-time acquisition of raw acoustic emission signals during bearing operation; S200: The CEEMDAN algorithm is used to decompose the original acoustic emission signal into multiple IMF components; S300: Fault characterization factors are constructed based on the fusion of kurtosis and Pearson correlation coefficient, and key IMF components are screened out; S400: Calculate the permutation entropy of key IMF components, and construct the bearing state feature vector after normalization; S500: The DSSA algorithm is used to optimize the support vector machine classification model; S600: Input the bearing state feature vector into the optimized support vector machine classification model and output the air conditioner bearing fault type identification result.

[0031] In a specific embodiment, the air conditioner bearing fault diagnosis method includes: ①. Real-time acquisition of raw acoustic emission signals during bearing operation. Acoustic emission sensors are placed near the bearing housing of the air conditioner compressor or fan. High-frequency acoustic emission signals during bearing operation are acquired in real time through an acoustic emission acquisition system. The acquired high-frequency acoustic emission signals are used as the raw data input for subsequent analysis (i.e., raw acoustic emission signals); ②. Decomposition of the raw acoustic emission signals using the CEEMDAN algorithm to obtain multiple intrinsic mode function (IMF) components and residual signals. Specifically, the CEEMDAN algorithm effectively reduces mode aliasing and improves signal decomposition stability by introducing adaptive white noise and reconstructing it at each decomposition stage. It is suitable for the non-stationary characteristics of air conditioner bearing acoustic emission signals. Analysis; ③. Since not all IMF components contain valid fault information, a kurtosis-correlation coefficient fusion screening method is used to screen the IMF components, selecting key IMF components; ④. The permutation entropy value is calculated for each of the selected key IMF components and normalized to form a feature vector representing the operating state of the air conditioning bearing (i.e., the bearing state feature vector), which is used as input to the support vector machine classification model; ⑤. An improved diversity adaptive sparrow search algorithm (DSSA) is used to optimize the penalty parameters and kernel function parameters of the support vector machine to improve classification accuracy; ⑥. The constructed bearing state feature vector is input into the SVM classification model optimized by DSSA to realize the identification of normal state, inner ring fault, outer ring fault and rolling element fault of the air conditioning bearing.

[0032] To address the non-stationary and high-noise characteristics of acoustic emission signals from air conditioning bearings, this application employs a CEEMDAN-based multi-scale decomposition method to process the signals. By fusing kurtosis features with Pearson correlation coefficients, an adaptive fault characterization mechanism for screening key modal components is constructed. Based on this, the permutation entropy method is used to construct fault features that reflect differences in bearing operating states. Furthermore, an improved diversity adaptive sparrow search algorithm is introduced to optimize the parameters of the support vector machine classification model, thereby forming a complete technical solution for acoustic emission fault diagnosis of air conditioning bearings.

[0033] Further, see Figure 3 The DSSA algorithm includes: introducing a population diversity preservation strategy based on Euclidean distance to perform population diversity initialization; introducing a Gaussian perturbation strategy to update the discoverer's position by adaptively adjusting the perturbation amplitude; and introducing a fitness difference-guided strategy based on global optimum to update the joiner's position.

[0034] Specifically, the DSSA algorithm is a diversity adaptive sparrow search algorithm. The sparrow population mainly consists of three types of individuals: discoverers, followers, and scouts. The sparrow search algorithm performs global optimization by simulating the foraging behavior and danger perception mechanism of sparrows. However, the traditional sparrow search algorithm (SSA) has defects such as uneven initial population distribution, slow later convergence speed, easy premature convergence, and insufficient global optimization ability, and cannot meet the parameter optimization requirements under high-dimensional, multi-peak, and strong noise conditions. To improve this situation, this application uses multiple strategies to improve the traditional sparrow algorithm: ①. The initialization of population diversity uses the Euclidean distance as the basis for judging population uniformity, which is used to construct a uniformly distributed initial sparrow population, providing a stable solution space basis for the subsequent global search of discoverers and the position tracking of joiners. And through this initialization step, the population has sufficient global search potential, providing a diverse initial direction for the position update of discoverers; ②. The position update of discoverers uses Gaussian perturbation as the core strategy,承接上述多样性初始种群,用于提升算法全局搜索能力,为后续加入者位置引导提供更优的全局搜索方向;经高斯扰动更新后的发现者位置,能够有效跳出局部最优,为加入者提供更准确的全局最优导向,使发现者搜索与加入者追踪形成强关联的递进搜索机制;③. The position update of joiners uses the fitness difference as the basis for dynamically adjusting the step size, directly承接发现者提供的全局搜索方向,并以全局最优解为收敛目标,实现搜索方向与搜索步长的联动控制。

[0035] Further, a Gaussian perturbation strategy is introduced to update the position of the discoverer. By adaptively adjusting the perturbation amplitude, including: introducing a Gaussian distribution perturbation term to adaptively adjust the position of the discoverer: ; The perturbation amplitude is dynamically adjusted with the number of iterations: ; Among them, is the solution position of individual i after update, is the current solution position of individual i, represents the standard Gaussian distribution, that is, the mean is 0 and the standard deviation is of the perturbation; is the current global optimal solution, is the current worst solution, is; is the perturbation amplitude of the current iteration, is the initial perturbation amplitude, iter is the current iteration number, is the maximum number of iterations.

[0036] Specifically, in the traditional SSA algorithm, the discoverer individuals are mainly responsible for global exploration. This application introduces perturbation based on the Gaussian distribution, improves the search diversity when updating the discoverer individuals, and realizes the exploration-exploitation balance at different stages by adaptively adjusting the perturbation amplitude: ①. The probability density function of the Gaussian distribution is: It should be noted that there are some parts in the original text that seem to be incomplete or have unclear expressions, which may affect the accuracy of the translation. Please check and correct the original text if possible. ,in, Standard deviation, The variance is; ②. In the improved Gaussian perturbation strategy, a Gaussian distributed perturbation term is introduced to replace the location update method for individual discoverers: ③. The Gaussian perturbation amplitude is dynamically adjusted with each iteration. Specifically, the perturbation amplitude adaptively decreases with the number of iterations. A larger perturbation amplitude is used for broad searching in the early stages of the search, while a smaller perturbation amplitude is used for fine-tuning the search and enhancing local search capabilities in the later stages. The perturbation amplitude is adjusted with the number of iterations, as shown in the formula: .

[0037] The discoverer position update uses Gaussian perturbation as its core strategy, taking into account the diversity of the initial population to improve the algorithm's global search capability and provide a better global search direction for guiding the position of subsequent joiners. The discoverer position updated by Gaussian perturbation can effectively escape local optima and provide more accurate global optimum guidance for joiners, thus forming a strongly correlated progressive search mechanism between discoverer search and joiner tracking.

[0038] Furthermore, a population diversity preservation strategy based on Euclidean distance is introduced to perform population diversity initialization, including: calculating the Euclidean distance between any two individuals in the sparrow population; and removing individuals in the sparrow population whose Euclidean distance is less than a set threshold to obtain a uniform population.

[0039] Specifically, the optimization performance of the traditional SSA algorithm is affected by population diversity, especially in the initial distribution stage, which may lead to individuals being overly concentrated in the solution space, reducing the population's exploration efficiency. To enhance global search capabilities and avoid premature convergence, this paper proposes a population diversity preservation strategy based on Euclidean distance, thereby constructing a uniformly distributed sparrow population in the solution space. In the original SSA, the population individuals are randomly initialized, which may lead to individuals being overly concentrated in the solution space, reducing the population's exploration efficiency. The specific steps include: ①. Calculating the Euclidean distance between sparrow population individuals: ,in, and They are two individuals from a sparrow population. From the perspective of the problem dimension, For individuals and individuals ① The Euclidean distance between two individuals; ②. When the Euclidean distance between two individuals... At this time, any one of the individuals is removed to ensure population diversity. By adopting a population diversity preservation strategy based on Euclidean distance, the solution space coverage capability of the SSA algorithm in the initialization phase can be significantly enhanced, laying a good foundation for the subsequent optimization process.

[0040] Furthermore, a fitness difference-based strategy is introduced to update the joiner's position, including: calculating the fitness difference between the current individual and the global optimum. Dynamically adjust the search step size by combining fitness differences and the global optimal solution: Update the joiner positions with the goal of achieving the global optimal solution: .in, For fitness differences, For individuals fitness The fitness of the global optimal solution; Let j be the search step size. This is the step size adjustment factor. These are uniformly distributed random numbers; The solution position for individual j after the update. For individuals The current solution position, This represents the location of the globally optimal solution.

[0041] Specifically, the joiner update strategy of the traditional SSA algorithm mainly relies on local information (such as the position of the optimal solution), which does not make sufficient use of the global optimal solution and limits the development capability. Therefore, this application proposes a fitness difference guidance strategy based on the global optimal solution to enhance the influence of the global optimal solution and improve the convergence speed and development capability.

[0042] When the fitness of an individual differs significantly from the global optimum, the individual moves rapidly toward the global optimum, accelerating the search: In the early stages of the algorithm, a large fitness difference promotes extensive exploration, quickly approaching the global optimum and enhancing global search capabilities; as iterations progress, the fitness difference decreases, the step size shortens, and the individual gradually converges to the global optimum, improving the accuracy of local exploration and achieving efficient and precise optimization; this strategy balances global exploration and local exploration, helping the algorithm accelerate convergence and avoid local optimum traps. The fitness difference formula is as follows: ,in, For fitness differences, For individuals fitness The fitness of the global optimum is used as the basis for determining the fitness difference and the global optimum. The search step size is then dynamically adjusted based on the fitness difference and the global optimum. ,in, Let j be the search step size. This is the step size adjustment factor. The numbers are uniformly distributed random numbers; the step size is guided by fitness differences, and the individual... The update formula is: ,in, The solution position for individual j after the update. For individuals The current solution position, This represents the location of the globally optimal solution.

[0043] Furthermore, the S500 also includes: Using SVM classification accuracy as the fitness function, the optimal penalty parameters and kernel function parameters are obtained through iterative optimization using the DSSA algorithm. With the goal of maximizing classification accuracy, output the optimal combination of SVM parameters; The support vector machine classification model is optimized by combining the optimal parameters of the SVM.

[0044] Specifically, the parameters of the Support Vector Machine (SVM) classification model include the penalty parameter C and the kernel function parameter γ. The classification performance of SVM is highly dependent on the penalty parameter C and the kernel function parameter γ. Traditional empirical settings, grid search, and simple optimization algorithms are prone to getting trapped in local optima, resulting in poor model generalization ability. After updating the discoverer's and joiner's positions, the algorithm enters the global optimal sparrow position and fitness update stage. This stage inherits all the steps of population initialization, discoverer update, and joiner update, and is used to record the optimal solution after each iteration and determine whether the maximum number of iterations has been reached. If the maximum number of iterations has not been reached, it returns to the discoverer position update step to continue iterating, forming a closed-loop iterative mechanism of "initialization - discoverer search - joiner guidance - global optimal update". If the maximum number of iterations has been reached, the final global optimal parameters are output as the optimal penalty parameters and kernel function parameters of the SVM model. In this application, the fitness function of the DSSA algorithm directly adopts the SVM classification accuracy, making the four core elements of population diversity, Gaussian perturbation, fitness difference, and global optimal solution strongly correlated with fault classification accuracy. DSSA continuously optimizes the SVM parameters through the above progressive correlation steps, enabling the SVM model to have higher classification accuracy and stronger generalization ability when facing the nonlinear feature vector of the acoustic emission signal of the air conditioner bearing, and finally realizing the complete technical link of feature vector - optimization algorithm - classification model - fault identification result.

[0045] The DSSA algorithm is based on population diversity initialization, with discoverer position update as the core of global search, joiner position update as the local development guide, and global optimal solution and fitness update as the iterative convergence goal. The four links are sequentially related and progressively advance to form a complete parameter optimization closed loop.

[0046] Furthermore, the S200 includes: S210: Adaptive white noise is added to the original acoustic emission signal to construct an auxiliary signal; S220: Perform multiple EMD decompositions on the auxiliary signal and take the average value to obtain the first-order IMF component; S230: Calculate the first-order residual component based on the first-order IMF component, and continue to add white noise for iterative decomposition to obtain the IMF components and residual components of each order until the residual component is a monotonic indivisible signal. S240: Represent the original acoustic emission signal as the sum of the IMF components of each order and the final residual component.

[0047] Specifically, traditional empirical mode decomposition (EMD) and ensemble empirical mode decomposition (EEMD) suffer from severe mode aliasing when processing non-stationary signals, resulting in component distortion, fault information being submerged by noise, and unreliable decomposition results.

[0048] The specific steps of the CEEMDAN algorithm in this application are as follows: ①. After adding adaptive white noise, a new auxiliary signal is constructed: ,in, This is the auxiliary signal for the i-th test. The original acoustic emission signal contains white noise. The white noise sequence for the i-th test. The standard deviation of the white noise sequence. ① Number of trials; ②. Auxiliary signal after adding white noise EMD decomposition yields a set of intrinsic mode functions. Averaging these mode components yields the first-order IMF component of the CEEMDAN decomposition. Calculate the first-order residual components based on the first-order IMF components: ③. White noise Add first-order residual components In the middle, to carry out This is a series of trials; in each trial, the signal... The process continues until the components are reduced to first-order components, yielding second-order IMF components: Calculate the second residual component: Repeat the above steps until the residual signal is a monotonic function and cannot be further decomposed; ④. At this point, we obtain The final residual signal consists of several IMF components. ,Signal It can be represented as: ,in, The image shows the original acoustic emission signal after decomposition and updating using the CEEMDAN algorithm. The CEEMDAN algorithm in this application introduces an adaptive white noise-assisted decomposition process. This strategy can significantly reduce mode aliasing in EMD processing. Furthermore, the CEEMDAN algorithm also exhibits small reconstruction error and high decomposition efficiency.

[0049] Furthermore, the S300 includes: S310: Calculate the kurtosis value of each IMF component; S320: Calculate the Pearson correlation coefficient between each IMF component and the original acoustic emission signal; S330: After normalizing the kurtosis value and Pearson correlation coefficient respectively, the values ​​are weighted and fused to construct the fault characterization factor; S340: Sort the fault characterization factors from largest to smallest and select the top m IMF components as key IMF components.

[0050] Specifically, for bearing acoustic emission signals, when the bearing is operating normally, the probability distribution of the signal approximates a normal distribution, and periodic impacts are not significant. However, when the bearing fails, due to the periodic collisions between the rolling elements and the fault point, the signal probability distribution tends to be steeper, and the kurtosis value increases significantly. Kurtosis is used to measure the intensity of transient impacts and is expressed as: Where x is the original acoustic emission signal, The mean of the original acoustic emission signal. The standard deviation of the original acoustic emission signal is given; however, in addition to containing fault information, the bearing acoustic emission signal is also superimposed with background noise generated by the operation of the mechanical system and vibration interference from other components.

[0051] Therefore, this application employs a standardization method to process the kurtosis values ​​of the IMF components in order to reduce the influence of the original signal kurtosis: ,in, For normalized kurtosis values, Let be the kurtosis value of the k-th IMF component. This represents the kurtosis value of the original acoustic emission signal.

[0052] The Pearson correlation coefficient is used to measure the correlation between the IMF component and the original signal. Its calculation formula is as follows: ,in, For the first One IMF component, This is the original acoustic emission signal.

[0053] Acoustic emission (IMF) fault information in rolling bearings mainly manifests as transient impacts. Due to the complexity of operating conditions and the influence of propagation paths, the selection of IMF components often relies on manual experience, fixed thresholds, or single evaluation indicators, making it difficult to comprehensively characterize the fault features within the IMF components. Kujicic can capture short-term impact characteristics but is susceptible to random impulse noise interference; correlation coefficients can reflect the global trend of the signal but are insensitive to weak fault impacts and cannot adaptively extract components that truly contain effective fault information. To overcome the limitations of a single criterion, this paper combines kujicic and correlation coefficients to construct a multidimensional fault characterization factor P (hereinafter referred to as the fault factor), defined as follows: ,in, For normalized kurtosis values, This refers to the Pearson correlation coefficient. This method allows us to identify the component that best reflects the fault characteristics of a signal across multiple modal components, avoiding the blindness of relying on a single threshold or fixed standard.

[0054] Preferably, the permutation entropy is calculated using the following formula: ,in, It is the first The probability of a sorting pattern occurring, where r is the total number of sorting patterns.

[0055] Furthermore, the types of air conditioning bearing failures include: normal condition, inner ring failure, outer ring failure, and rolling element failure.

[0056] Specifically, by classifying the types of air conditioner bearing failures, fault location and classification can be achieved, guiding precise repairs.

[0057]

Example 2

[0058] In one specific embodiment, the acquisition module 101, decomposition module 102, screening module 103, permutation entropy feature module 104, optimization module 105 and identification module 106 of the air conditioner bearing fault diagnosis system 100 work together to implement the air conditioner bearing fault diagnosis method based on acoustic emission signals as described in the first embodiment above, which will not be elaborated here.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing air conditioner bearing faults based on acoustic emission signals, characterized in that, The air conditioner bearing fault diagnosis method includes: Real-time acquisition of raw acoustic emission signals during bearing operation; The original acoustic emission signal is decomposed into multiple IMF components using the CEEMDAN algorithm. Fault characterization factors were constructed by fusing kurtosis and Pearson correlation coefficient, and key IMF components were screened out. The permutation entropy of the key IMF components is calculated and normalized to construct the bearing state feature vector; The DSSA algorithm is used to optimize the support vector machine classification model; The bearing state feature vector is input into the optimized support vector machine classification model, and the air conditioner bearing fault type identification result is output.

2. The air conditioner bearing fault diagnosis method according to claim 1, characterized in that, The DSSA algorithm includes: A population diversity preservation strategy based on Euclidean distance is introduced to perform population diversity initialization; A Gaussian perturbation strategy is introduced to update the discoverer's position, with the perturbation amplitude being adaptively adjusted. A fitness difference-based strategy based on global optimum is introduced to update the joiner's position.

3. The air conditioner bearing fault diagnosis method according to claim 2, characterized in that, The method of introducing a Gaussian perturbation strategy to update the discoverer's position, by adaptively adjusting the perturbation amplitude, includes: The location of the discoverer is adaptively adjusted by introducing a Gaussian distribution perturbation term: ; The perturbation amplitude is dynamically adjusted with the number of iterations: ; in, The solution position for individual i after the update. Let i be the current solution position of individual i. This represents a standard Gaussian distribution, with a mean of 0 and a standard deviation of . The disturbance; This is the current globally optimal solution. This is the worst solution at present; The perturbation magnitude in the current iteration, The initial perturbation magnitude is given by `iter`, where `iter` is the current iteration number. This represents the maximum number of iterations.

4. The air conditioner bearing fault diagnosis method according to claim 2, characterized in that, The proposed population diversity preservation strategy based on Euclidean distance, which performs population diversity initialization, includes: Calculate the Euclidean distance between any two individuals in a sparrow population; Individuals in the sparrow population whose Euclidean distance is less than a set threshold are removed to obtain a uniform population.

5. The air conditioner bearing fault diagnosis method according to claim 2, characterized in that, The introduction of a fitness difference-guided strategy based on global optimum for updating joiner positions includes: Calculate the fitness difference between the current individual and the global optimum: ; The search step size is dynamically adjusted based on the fitness differences and the global optimal solution: ; Update the joiner positions with the global optimal solution as the target: ; in, For individuals Differences in fitness For individuals fitness The fitness of the global optimal solution; Let j be the search step size. This is the step size adjustment factor. These are uniformly distributed random numbers; The solution position for individual j after the update. For individuals The current solution position, This represents the position of the global optimal solution.

6. The air conditioner bearing fault diagnosis method according to claim 2, characterized in that, The optimization of the support vector machine classification model using the DSSA algorithm also includes: Using the SVM classification accuracy as the fitness function, the optimal penalty parameters and kernel function parameters are obtained through iterative optimization using the DSSA algorithm. With the goal of maximizing classification accuracy, output the optimal combination of SVM parameters; The support vector machine classification model is optimized using the optimal combination of SVM parameters.

7. The air conditioner bearing fault diagnosis method according to claim 1, characterized in that, The original acoustic emission signal is decomposed into multiple IMF components using the CEEMDAN algorithm, including: Adaptive white noise is added to the original acoustic emission signal to construct an auxiliary signal; The auxiliary signal is subjected to multiple EMD decompositions and the average value is taken to obtain the first-order IMF component; The first-order residual component is calculated based on the first-order IMF component, and white noise is added to perform iterative decomposition to obtain the IMF components and residual components of each order until the residual component is a monotonically indivisible signal. The original acoustic emission signal is represented as the sum of each order IMF component and the final residual component.

8. The air conditioner bearing fault diagnosis method according to claim 1, characterized in that, The fault characterization factor, constructed by fusing kurtosis and Pearson correlation coefficient, is used to screen out key IMF components, including: Calculate the kurtosis value of each of the IMF components; Calculate the Pearson correlation coefficient between each of the IMF components and the original acoustic emission signal; The fault characterization factor is constructed by normalizing the kurtosis value and the Pearson correlation coefficient respectively and then weighting and fusing them. The first m IMF components are selected as the key IMF components by sorting the fault characterization factors from largest to smallest.

9. The air conditioner bearing fault diagnosis method according to claim 1, characterized in that, The types of air conditioning bearing failures include: normal condition, inner ring failure, outer ring failure, and rolling element failure.

10. A fault diagnosis system for air conditioning bearings based on acoustic emission signals, characterized in that, The air conditioner bearing fault diagnosis system adopts the air conditioner bearing fault diagnosis method based on acoustic emission signals as described in any one of claims 1-9, and the air conditioner bearing fault diagnosis system includes: Acquisition module, which is used to acquire the raw acoustic emission signals during the bearing operation process in real time; The decomposition module is used to decompose the original acoustic emission signal into multiple IMF components using the CEEMDAN algorithm. The screening module is used to construct fault characterization factors based on the fusion of kurtosis and Pearson correlation coefficient, and to screen out key IMF components; The permutation entropy feature module is used to calculate the permutation entropy of the key IMF components and construct the bearing state feature vector after normalization. The optimization module is used to optimize the support vector machine classification model using the DSSA algorithm; The identification module is used to input the bearing state feature vector into the optimized support vector machine classification model and output the air conditioner bearing fault type identification result.

Images