An artificial intelligence-based fan safety fault detection method

Abstract

The present application belongs to the technical field of fault detection, and particularly relates to a fan safety fault detection method based on artificial intelligence. The method first collects original signal samples, and then carries out pretreatment such as denoising and filtering. Then, the improved artificial bee colony algorithm ABC is used to optimize feature modal decomposition FMD, and the intrinsic modal function IMF is obtained by screening local extreme values. The optimization problem is constructed with the minimum error function and the maximum information gain as the target, and a series of operations are carried out to obtain the signal modal component. Finally, the KDTree considering information gain is used to divide the dimension to construct the tree according to the data characteristics, and the signal component data is included in the tree, and the fault judgment is realized by traversing the leaf node. The present application effectively solves the problems of low efficiency and poor accuracy of the existing fan fault detection method, improves the accuracy and efficiency of fan fault detection, and ensures the safe operation of the fan.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault detection, and particularly relates to a fan safety fault detection method based on artificial intelligence. BACKGROUND

[0002] In industrial production and daily life, fans are widely used as common heat dissipation equipment. However, various safety faults may occur in the long-term operation of the fan, such as motor failure, blade damage, bearing wear, etc. These faults not only affect the normal operation of the fan and reduce the heat dissipation effect, but also may cause safety accidents, such as equipment overheating damage, fire, etc. The existing fan fault detection methods mainly rely on manual detection and simple sensor monitoring. This way is low in efficiency and difficult to guarantee accuracy, and often can only find problems after the fault occurs, which cannot realize early warning and accurate prediction of faults. SUMMARY

[0003] The application proposes a fan safety fault detection method based on artificial intelligence to solve the technical problems in the above background technology.

[0004] In order to achieve the above purpose, the technical scheme adopted by the application comprises the following steps:

[0005] S1, first collect original signal samples;

[0006] S2, pre-process the original signal;

[0007] S3, use the improved artificial bee colony algorithm ABC to optimize feature modal decomposition FMD to realize maximum extraction of fault feature information accuracy and minimum error between extracted signal components and real fault feature signals, and extract signal components of fault information from the original signal, and the specific implementation steps are as follows:

[0008] S31, first obtain the intrinsic modal function IMF by gradually screening the local maximum and minimum values of the signal: wherein is the original signal, is the i-th intrinsic modal function, is the residual signal;

[0009] S32, determine the optimization target as the minimum error function: wherein is the real fault signal corresponding to the modal component;

[0010] S33, use information gain to measure the contribution of each modal component to the fault, and the information gain is expressed as: wherein is the feature of the modal component, Entropy of failure, is a known feature Conditional entropy of failure under the condition;

[0011] S34, finally, a new optimization problem is constructed to control the error minimization and information gain maximization: Where N is the number of modal components, Q is the number of features corresponding to each modal component, improved ABC optimization FMD is adopted, and a group of signal modal components is obtained by ABC optimization;

[0012] S4, finally, KDTree based on considering information gain is adopted, the information gain is calculated according to the data characteristics to select the division dimension to build the tree, and the signal component data is included in the tree to realize the fault judgment according to the leaf node.

[0013] As preferred, the original signal sample in step S1 includes vibration signal, temperature signal, current signal and audio signal.

[0014] As preferred, the operation of pre-processing the original signal in step S2 includes denoising, filtering, standardization and smoothing operation.

[0015] As preferred, the specific operation of improving ABC algorithm to realize optimization in step S34 is:

[0016] S351, first, initialize and randomly generate the positions of P bees, representing potential solutions, Wherein represents the position of the jth bee, and m is the dimension of the solution; Set the initial search range Wherein are the minimum and maximum boundaries respectively, and for each solution Calculate its fitness value; ;

[0017] S352, adjust the search range according to the fitness change, if the current fitness Is less than the last generation, then narrow the search range: Wherein is the proportion of controlling the range narrowing, which is 0.1, if the current fitness Is greater than the last generation, then expand the search range: Wherein are the search ranges before and after adjustment respectively;

[0018] S353, then the bees start searching, each hired bee searches in the new search range, and the search formula is: Wherein is a random factor between [-1, 1], which controls the step length of search;

[0019] S354, after each search, update the fitness function and evaluate the quality of the current solution, select the solution with the optimal fitness as the global optimal solution;

[0020] S355, continue iterative optimization until the predetermined maximum number of iterations of 50 times is reached to terminate the iteration.

[0021] As preferred, in order to avoid too small search range in step S352, a minimum threshold is set : to prevent excessive narrowing of the search range, resulting in ineffective search of new solution space.

[0022] As preferred, the specific implementation of the fault judgment in step S4 is:

[0023] S41, first select the optimal feature according to information gain to construct the root node of the tree;

[0024] S42, according to the value of the selected feature, select the median of the feature to divide the data into two parts, the left subtree corresponds to , and the right subtree corresponds to ;

[0025] S43, recursively perform the same operation on the left and right subtrees, and select the feature with the maximum information gain for division;

[0026] S44, starting from the root node of the tree, when traversing to the leaf node, the leaf node is marked as fault or non-fault, that is, the fault judgment result of the output signal component.

[0027] Compared with the prior art, the advantages and positive effects of the present application are that multiple original signals are comprehensively monitored to avoid the limitations of single signal monitoring. The improved ABC algorithm is used to optimize FMD to improve the accuracy of fault feature extraction. The improved algorithm can better balance the search ability and obtain better signal modal components. The KDTree considering information gain is used for fault judgment to quickly and accurately determine the fault. The features are selected to construct the tree, and the data is recursively divided to efficiently mine fault features, realize early warning and accurate prediction of fan faults, and ensure safe and stable operation of equipment. DETAILED DESCRIPTION

[0028] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0029] ​Many specific details are set forth in the following description in order to provide a thorough understanding of the application. However, the application can be practiced according to the claims without resorting to the many specific details set forth in the following description.

[0030] In the embodiment, the conventional fan fault detection method has many problems. For example, manual detection and simple sensor monitoring are inefficient, inaccurate, and cannot provide early warning of faults. In order to solve these problems, the application proposes a fan safety fault detection method based on artificial intelligence, aiming to improve the accuracy and efficiency of fan fault detection, realize early warning, and ensure the safe and stable operation of the fan.

[0031] The collected original signal samples include vibration signals, temperature signals, current signals, and audio signals. The vibration signal can reflect the balance state of the fan blade, the wear condition of the bearing, etc.; the temperature signal can indicate whether the motor is overheating, thereby indirectly reflecting the working state of the motor; the change of the current signal is related to the load of the motor, winding fault, etc.; the audio signal can capture whether there is abnormal noise during the operation of the fan, such as blade scraping sound, motor abnormal sound, etc. By comprehensively collecting these different types of signals, the running condition of the fan can be monitored from multiple angles, avoiding the missed detection or misjudgment problem caused by single signal monitoring. In order to improve the signal quality and facilitate subsequent accurate analysis, preprocessing operations are implemented. Including denoising, filtering, standardization and smoothing operations on the original signal.

[0032] In order to maximize the accuracy of extracting fault feature information and minimize errors, an improved artificial bee colony algorithm ABC is used to optimize feature modal decomposition FMD to accurately select the local maximum and minimum values of the signal to obtain the intrinsic modal function IMF. The complex signal is decomposed into multiple meaningful components to deeply mine the internal structure. The optimization target is clear, and by minimizing the error function, the extracted signal component is approximated to the real fault signal, improving the accuracy of fault feature extraction. The information gain is used to measure the contribution of the modal component to the fault, and the key component is selected. The improved ABC algorithm balances global and local search to avoid falling into local optimum, thereby more effectively obtaining the optimal signal modal component. The specific implementation is as follows: first, the local maximum and minimum values of the signal are selected step by step to obtain the intrinsic modal function IMF: wherein is the original signal, is the i-th intrinsic modal function, is the residual signal; the optimization target is determined as minimizing the error function: wherein is the real fault signal corresponding to the modal component; the information gain is used to measure the contribution of each modal component to the fault, and the information gain is represented as: wherein is the feature of the modal component, Entropy of failure, is the known feature Conditional entropy of failure under the condition; Finally, a new optimization problem is constructed to control the minimum error and maximum information gain: , where N is the number of modal components, Q is the number of features corresponding to each modal component, and the improved ABC optimization FMD is used to obtain a set of signal modal components.

[0033] The improved ABC algorithm is used to optimize the first initialization, randomly generating P positions of bees, representing potential solutions, , where represents the position of the jth bee, and m is the dimension of the solution; Set the initial search range , where are the minimum and maximum boundaries, respectively, and for each solution , calculate its fitness value; ; according to the fitness change, adjust the search range, if the current fitness is less than the last generation, then narrow the search range: , where is the control range narrowing ratio, which is 0.1, if the current fitness is greater than the last generation, then expand the search range: , where are the search ranges before and after adjustment, respectively; then the bees start searching, each hired bee searches in the new search range, the search formula is: , where is a random factor between [-1, 1], which controls the step size of the search; after each search, update the fitness function and evaluate the quality of the current solution, select the solution with the best fitness as the global optimal solution; continue iteration optimization until the maximum iteration number of 50 is reached to terminate the iteration.

[0034] Finally, in order to accurately judge the fan failure, KDTree based on considering information gain is used to calculate information gain according to data features to select the division dimension to build the tree, which can preferentially use the most discriminant feature as the node of the tree, making the data classification more efficient and accurate. When processing fan failure data, by recursively selecting the feature with the maximum information gain, the classification can be gradually refined, and the fault features can be accurately mined. From the root node to the leaf node, the fault judgment can be realized, which can quickly give the judgment result of failure or non-failure, effectively improving the efficiency and accuracy of fault diagnosis. First, the optimal feature is selected according to the information gain to construct the root node of the tree; according to the value of the selected feature, the median of the feature is selected to divide the data into two parts, the left subtree corresponds to , and the right subtree corresponds to ; recursively the same operation is performed on the left and right sub-trees, selecting the feature with the largest information gain to split; finally, starting from the root node of the tree, when a leaf node is reached, it is labeled as either faulty or non-faulty, i.e. the fault decision result of the output signal component.

[0035] The above merely describes preferred embodiments of the present application, but is not intended to limit the present application in other forms. Any person skilled in the art can modify or change the above disclosed technical contents into equivalent embodiments with equivalent changes, and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical scheme of the present application, still falls within the protection scope of the present application.

Claims

1. An artificial intelligence-based fan safety fault detection method, characterized by, The method comprises the following steps: S1, first collecting original signal samples; S2, pre-processing the original signal; S3, using an improved artificial bee colony algorithm ABC to optimize feature mode decomposition FMD to maximize the accuracy of extracted fault feature information and minimize the error between extracted signal components and real fault feature signals, and extracting signal components of fault information from the original signal, the specific implementation steps are: S31, first, by step by step screening signal local maximum and minimum, get intrinsic mode function IMF: wherein is the original signal, is the i-th intrinsic mode function, is the residual signal; S32, determine the optimization target as minimizing the error function: wherein is the real fault signal corresponding to the modal component; S33, contribution of each modal component to the fault is measured by information gain, which is expressed as: wherein is the feature of the modal component, is the entropy of the fault, is the conditional entropy of the fault under the condition that the feature is known. S34, finally, a new optimization problem is constructed to control the error minimization and information gain maximization: where N is the number of modal components, Q is the number of characteristics corresponding to each modal component, and the improved ABC optimization FMD is used to obtain a set of signal modal components. S4, finally, based on KDTree considering information gain, the tree is constructed according to the data feature calculation information gain to select the partition dimension, and the signal component data is included in the tree to realize the fault judgment according to the leaf node.

2. The method of claim 1, wherein the method is based on artificial intelligence. The original signal samples in step S1 include vibration signals, temperature signals, current signals and audio signals.

3. The method of claim 1, wherein the method is based on artificial intelligence. The pre-processing operation of the original signal in step S2 includes denoising, filtering, standardization and smoothing operation.

4. The method of claim 1, wherein the method is based on artificial intelligence. The specific operation of the improved ABC algorithm in step S34 to realize optimization is: S351、First, initialization, randomly generate the position of P bees, representing potential solutions, where represents the position of the jth bee, m is the dimension of the solution; set the range of initial search where are the minimum and maximum boundaries, respectively, and for each solution calculate its fitness value; ; S352, adjust the search range according to the fitness change, if the current fitness is less than the last generation, then narrow the search range: wherein is the proportion of the control range narrowing, is 0.1, if the current fitness is greater than the last generation, then expand the search range: wherein are the search ranges before and after adjustment, respectively; S353、Then the bees start searching, each employed bee in a new search range, the search formula is: where is a random factor between [-1, 1] that controls the step size of the search; S354, after each search, update the fitness function and evaluate the quality of the current solution, and select the solution with the optimal fitness as the global optimal solution; S355, continue to iterate and optimize until the predetermined maximum iteration number of 50 is reached to terminate the iteration.

5. The method of claim 4, wherein the method further comprises: In order to avoid too small search range, a minimum threshold is set in the step S352 : , which prevents over-narrowing the search range and leads to ineffective search of new solution space.

6. The method of claim 1, wherein the method is based on artificial intelligence. The specific implementation of realizing the fault judgment in step S4 is: S41, first select optimal features according to information gain to construct the root node of the tree; S42, select the median of the feature according to the value of the selected feature The data is divided into two parts, the left subtree corresponds to , the right subtree corresponds to ; S43, recursively perform the same operation on the left and right subtrees, and select the feature with the maximum information gain for partitioning; S44, starting from the root node of the tree, when a leaf node is reached, the leaf node is marked as fault or non-fault, that is, the fault judgment result of the signal component is output.

Images