Abnormal sound detection method, device, equipment and medium

Abstract

The present invention provides an abnormal sound detection method, device, equipment and medium, including: obtaining industrial equipment sound signals, extracting multidimensional feature data therefrom; performing state mapping on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of different state mapping tables; calling a trained Markov model to perform probability calculation on the multidimensional feature state sequence to generate a feature state transition probability, performing comparison processing on the feature state transition probability to generate a comparison result, wherein the Markov model has a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and the feature state transition probability is compared with the feature state transition probability matrix group to obtain a comparison result; and outputting a warning signal according to the comparison result. The invention aims to solve the problem that the existing sound detection scheme cannot distinguish between abnormalities of different time series patterns, is prone to overfitting and lacks interpretability.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality detection, and in particular to an abnormal sound detection method, device, equipment and medium. Background Art

[0002] Abnormal sound detection has a wide range of application scenarios in the fields of industrial equipment monitoring, product automated quality inspection, etc., and there are many research results in anomaly detection technology, mainly in the two directions of machine learning and neural networks. In machine learning methods, features such as distance, density, angle and tree depth are often constructed, and a benchmark distribution is constructed based on the features of normal samples. Data with features deviating from the benchmark distribution are identified as abnormal. This type of method is easy to train and has strong interpretability, but it cannot effectively distinguish abnormalities from different time series patterns; in neural network methods, network parameters can be obtained by fitting the time series features of normal samples. If the fitting value is significantly different from the actual value during real-time detection, the data will be identified as abnormal. This method takes into account the differences in data time series, but requires a large amount of data for network training, and parameter optimization is complex. It is easy to overfit and lacks interpretability.

[0003] In actual industrial scenarios, when equipment is loose or blocked, the changes in sound signal characteristics caused by it have a more complex pattern. To detect abnormalities in equipment sound characteristics, it is necessary not only to compare them with the statistical distribution of benchmark data, but also to determine the abnormality level and type from the time series trend of the data. Therefore, based on the characteristics of industrial equipment sound signals, it is an urgent task to develop an abnormality detection method that has a more efficient training process, supports diversified time series anomaly detection, and is highly interpretable.

[0004] In view of this, this application is filed. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide an abnormal sound detection method, device, equipment and medium, which can effectively solve the problems of abnormal sound detection methods in the prior art that they cannot effectively distinguish anomalies from different time series patterns, require a large amount of data for network training, have complex parameter optimization, are prone to overfitting and lack interpretability.

[0006] The present invention discloses an abnormal sound detection method, comprising:

[0007] Acquire the sound signals of industrial equipment and extract multi-dimensional feature data from them;

[0008] Performing state mapping processing on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed in different division modes;

[0009] Calling the trained Markov model to perform probability calculation on the multidimensional feature state sequence to generate corresponding feature state transition probabilities, and performing comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and the feature state transition probability is compared with the feature state transition probability matrix group to obtain corresponding comparison results;

[0010] According to the comparison result, a warning signal is output.

[0011] Preferably, before performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence, the method further includes:

[0012] Acquire multiple groups of historical sound signals and extract corresponding multi-dimensional feature data therefrom;

[0013] Dividing the characteristic value interval of the multidimensional characteristic data according to a preset distance step length to generate a corresponding state mapping table;

[0014] A plurality of groups of the state mapping tables are combined to generate a state mapping table group.

[0015] Preferably, before calling the trained Markov model to perform probability calculation on the multi-dimensional feature state sequence and generating the corresponding feature state transition probability, the method further includes:

[0016] Acquire multiple groups of normal historical sound signals and extract corresponding multi-dimensional feature data from them;

[0017] Performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence;

[0018] Performing Markov hypothesis processing on the multidimensional characteristic state sequence, screening out multidimensional characteristic state sequences that are all Markov processes, generating corresponding characteristic state transition probabilities, and converting them into characteristic state transition probability matrices;

[0019] A plurality of groups of the characteristic state transition probability matrices are combined to generate a group of characteristic state transition probability matrices.

[0020] Preferably, the Markov hypothesis is that the random variable at a certain moment depends only on the random variable at the previous moment, that is, p(X t |X 0 ,X 1 ,...,X t-1 )=p(X t |X t-1 ), where t = 1, 2, ..., a random sequence that satisfies the Markov property is called a Markov process.

[0021] Preferably, the characteristic state transition probability is calculated according to the formula p ij = p(X t =i|X t-1 =j) is converted into the characteristic state transition probability matrix, where i = 1, 2, ...; j = 1, 2, ..., p ij ≥0,∑p ij =1.

[0022] The present invention also provides an abnormal sound detection device, comprising:

[0023] A data acquisition unit, used to acquire sound signals of industrial equipment and extract multi-dimensional feature data therefrom;

[0024] A data processing unit, used for performing state mapping processing on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed by using different division methods and different implementation functions;

[0025] A data comparison unit is used to call the trained Markov model to perform probability calculation on the multidimensional feature state sequence, generate corresponding feature state transition probabilities, and perform comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and compares the feature state transition probability with the feature state transition probability matrix group to obtain corresponding comparison results;

[0026] The early warning unit is used to output an early warning signal according to the comparison result.

[0027] The present invention also provides an abnormal sound detection device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, an abnormal sound detection method as described above is implemented.

[0028] The present invention also provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the storage medium is located to implement an abnormal sound detection method as described in any one of the above.

[0029] In summary, the abnormal sound detection method, device, equipment and medium provided in this embodiment extract the sound signals of industrial equipment collected from actual industrial scenes to obtain multi-dimensional feature data, and use a state mapping table group to perform state mapping processing on the obtained multi-dimensional feature data, generate a multi-dimensional feature state sequence as the input parameter of the Markov model, and issue an early warning based on the comparison result of the model output, thereby solving the problems that the abnormal sound detection method in the prior art cannot effectively distinguish anomalies from different time series patterns, requires a large amount of data for network training, and has complex parameter optimization, is prone to overfitting and lacks interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flowchart of an abnormal sound detection method provided by an embodiment of the present invention.

[0031] Figure 2 1 is a schematic diagram of a first effect waveform of the abnormal sound detection method provided by an embodiment of the present invention.

[0032] Figure 3 4 is a schematic diagram of a second effect waveform of the abnormal sound detection method provided by an embodiment of the present invention.

[0033] Figure 4 It is a module schematic diagram of the abnormal sound detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0035] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0036] See also Figure 1 The first embodiment of the present invention provides an abnormal sound detection method, which can be performed by an abnormal sound detection method device (hereinafter referred to as a detection device), in particular, by one or more processors in the detection device to implement the following steps:

[0037] S101, obtaining sound signals of industrial equipment and extracting multi-dimensional feature data therefrom;

[0038] Specifically, in this embodiment, the multidimensional feature data may include time domain features: short-time zero-crossing rate, silence ratio, frequency features: FFT data, SFFT data, MFCC data, and statistical features: statistics such as mean and standard deviation calculated based on time-frequency domain features. It should be noted that in other embodiments, other types of multidimensional feature data may also be used, which are not specifically limited here, but these solutions are within the protection scope of the present invention.

[0039] For example, taking the monitoring of a certain device in an industrial scenario as an example, the sound collector is deployed near the device, the sampling rate is configured to 48000hz, and the calculation frequency is 24hz. The low-frequency band meanlf and the full-frequency band mean of the single-channel sound signal are calculated as feature data and uploaded to the server. For the historical features on the server, the time interval during which the equipment is operating normally is screened out, and the feature data within the time interval with the low-frequency band mean feature meanlf greater than the threshold 5e9 is classified as the operation type using the threshold method, and the remaining data is classified as the shutdown type. For the real-time feature data of the sound signal of the operation type, the maximum value is calculated using a one-minute sliding window as the feature of the anomaly detection model.

[0040] S102, performing state mapping processing on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed in different division modes;

[0041] Specifically, in this embodiment, before performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence, the method further includes:

[0042] Acquire multiple groups of historical sound signals and extract corresponding multi-dimensional feature data therefrom;

[0043] Dividing the characteristic value interval of the multidimensional characteristic data according to a preset distance step length to generate a corresponding state mapping table;

[0044] A plurality of groups of the state mapping tables are combined to generate a state mapping table group.

[0045] In this embodiment, the plurality of groups of state mapping tables can be obtained by dividing the characteristic value interval into equidistant steps and assigning the state values ​​incrementally; the characteristic value interval can be divided into unequal steps and assigning the state values ​​incrementally; and other linear functions or nonlinear functions can be used for implementation. It should be noted that in other embodiments, other construction methods can also be used to construct the state mapping table, which is not specifically limited here, but these solutions are all within the protection scope of the present invention.

[0046] For example, taking Table 1 as an example, a state mapping table with unequal step lengths is used, and the full-band mean feature mean is mainly in the range of 1e8 to 1e10. The configured state mapping table is shown in Table 1:

[0047] Table 1

[0048]

[0049] S103, calling the trained Markov model to perform probability calculation on the multidimensional feature state sequence, generating corresponding feature state transition probabilities, and performing comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and the feature state transition probabilities are compared with the feature state transition probability matrix group to obtain corresponding comparison results;

[0050] Specifically, in this embodiment, before calling the trained Markov model to perform probability calculation on the multi-dimensional feature state sequence and generating the corresponding feature state transition probability, the method further includes:

[0051] Acquire multiple groups of normal historical sound signals and extract corresponding multi-dimensional feature data from them;

[0052] Performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence;

[0053] Performing Markov hypothesis processing on the multidimensional characteristic state sequence, screening out multidimensional characteristic state sequences that are all Markov processes, generating corresponding characteristic state transition probabilities, and converting them into characteristic state transition probability matrices;

[0054] A plurality of groups of the characteristic state transition probability matrices are combined to generate a group of characteristic state transition probability matrices.

[0055] Specifically, in this embodiment, the Markov hypothesis is that for a random variable at a certain moment, it only depends on the random variable at the previous moment, that is, p(X t |X 0 ,X 1 ,...,X t-1)=p(X t |X t-1 ), where t = 1, 2, ..., a random sequence that satisfies the Markov property is called a Markov process. It is assumed that the multidimensional feature state sequence is a Markov process and is a time-homogeneous Markov process, that is, the transition probability distribution is independent of time t. The feature state transition probability is calculated according to the formula p ij = p(X t =i|X t-1 =j) is converted into the characteristic state transition probability matrix, where i = 1, 2, ...; j = 1, 2, ..., p ij ≥0,∑p ij =1, in the same characteristic state sequence, the value set of the variable is the same and is defined in a finite discrete state space. The above probability can be expressed as a matrix

[0056] For example, taking Table 1 as an example, the state transfer probability matrix calculated by the full-band mean feature mean is shown in Table 2; it can be seen from Table 2 that the characteristic state and the transferred state are both within a small range, which is consistent with the fact that during the operation of industrial equipment, it often performs periodic motion with certain working parameters and the sound itself is relatively stable.

[0057] Table 2

[0058]

[0059] S104: outputting a warning signal according to the comparison result.

[0060] Specifically, in this embodiment, if the value of the real-time feature state transition probability in the state transition probability matrix group is lower than the threshold, an early warning is issued, for example, Among them, alarm is the comparison result, and threshold is the preset threshold; the abnormal sound detection method can not only realize the anomaly detection on the statistical distribution of sound features, but also effectively distinguish the time series anomalies of multi-mode data. It has low training complexity and strong interpretability, and can accurately and efficiently identify abnormal conditions in industrial production to ensure the production efficiency and safety of enterprises.

[0061] For example, taking the abnormal sound detection in the event of the ball falling off inside the bearing component of the above equipment as an example, the equipment is equipped with a two-level warning, and different levels of warnings and corresponding thresholds are as follows shown.

[0062] In this embodiment, Figure 2The characteristic data and corresponding status values ​​of the equipment from the 27th to the 29th of a certain month: meanlf-low-frequency mean characteristic, mean-full-frequency mean characteristic, operation-equipment operation type (0: shutdown, 1: operation), meanlf_status-low-frequency mean characteristic status, mean_status-full-frequency mean characteristic status; the low-frequency mean characteristic meanlf status and the full-frequency mean characteristic mean partial status are the status changes caused by the normal fluctuation of the characteristics during the operation of the equipment. The state where the full-frequency mean characteristic mean is higher than four is a special state caused by the change of characteristics after the abnormal sound appears when the equipment is abnormal.

[0063] in, Figure 3 The figure is a schematic diagram of the full-band mean feature mean triggering an early warning when an abnormal sound occurs: mean-full-band mean feature, operation-equipment operation type (0: shutdown, 1: operation), mean_status-full-band mean feature state, mean_prop-full-band mean feature state transition probability, alarm-abnormal sound early warning level; when the equipment operation type is shutdown, the feature state transition probability is fixed at 0.98, which is determined by Figure 3It can be seen that the sound characteristics of the equipment components began to become abnormal on the 27th, and the full-band mean characteristic mean fluctuated between state three and state four, with a state transition probability as low as 0.05. After 10:30 am on the 28th, the characteristics further transferred from state four to state five, and the corresponding state transition probability was lower than the first-level warning threshold of 0.0005, causing a first-level abnormal warning. Around 11:30 pm on the 28th, the full-band mean characteristic mean became more abnormal, and indirectly transferred to state six on the basis of state five, with a state transition probability lower than the second-level warning threshold of 0.0001, triggering a second-level abnormal warning. During the emergency shutdown, the full-band mean characteristic mean reached state seven; when the equipment was restarted for the first time after the shutdown maintenance, the full-band mean characteristic mean still had some fluctuations, and the equipment was restarted after another shutdown inspection, and it returned to normal working state; in this abnormal event, the on-site duty personnel detected the ball bearings on the processed materials through the metal detector at 12 o'clock at night on the 28th, and then stopped the machine urgently for abnormal investigation. Compared with on-site manual inspection, this method can start pushing abnormal warnings at 10:30 am on the 28th, about thirteen hours in advance, which can effectively prevent further component damage from causing equipment damage on a larger scale, extend the service life of the equipment, and ensure the production efficiency and safety of the enterprise. On the other hand, in existing industrial scenarios, there may be differences in the degree of attention for the timing anomalies in different directions of the characteristic state of the sound signal. For mechanical rotating equipment, the focus is often on whether there is an upward trend in the vibration characteristics of the sound signal. If the abnormal state is slightly higher than the normal state, it is necessary to pay close attention to the operation of the equipment and check whether there is abnormal interference. If the abnormal state is seriously higher than the normal state, it is necessary to shut down in time for maintenance and inspection; for water pump or pipeline type equipment, if the sound signal energy characteristics are weakened, that is, when the abnormal state is lower than the normal state, it is necessary to adjust the production plan or enable spare equipment and other measures to avoid equipment blockage affecting normal production order. For a variety of abnormal sound detection scenarios, when the abnormal state transfer of the sound feature occurs, the abnormal sound detection method can configure differentiated warning levels for different types of timing anomalies to meet the monitoring needs of actual scenarios.

[0064] See also Figure 4 The second embodiment of the present invention provides an abnormal sound detection device, comprising:

[0065] The data acquisition unit 201 is used to acquire the sound signal of the industrial equipment and extract multi-dimensional feature data therefrom;

[0066] The data processing unit 202 is used to perform state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed by using different partitioning methods and different implementation functions;

[0067] The data comparison unit 203 is used to call the trained Markov model to perform probability calculation on the multi-dimensional feature state sequence, generate corresponding feature state transition probabilities, and perform comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and compares the feature state transition probability with the feature state transition probability matrix group to obtain corresponding comparison results;

[0068] The warning unit 204 is used to output a warning signal according to the comparison result.

[0069] A third embodiment of the present invention provides an abnormal sound detection device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, an abnormal sound detection method as described above is implemented.

[0070] A fourth embodiment of the present invention provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the storage medium is located to implement an abnormal sound detection method as described in any one of the above.

[0071] Exemplarily, the computer program described in the third embodiment and the fourth embodiment of the present invention can be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the implementation of an abnormal sound detection device. For example, the device described in the second embodiment of the present invention.

[0072] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the abnormal sound detection method, and uses various interfaces and lines to connect the various parts of the abnormal sound detection method.

[0073] The memory can be used to store the computer program and / or module, and the processor realizes various functions of an abnormal sound detection method by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, a text conversion function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0074] Wherein, if the implemented module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0075] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0076] The above are only preferred implementations of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention.

Claims

1. A method for detecting abnormal sound, It is characterized in that include: Acquire the sound signals of industrial equipment and extract multi-dimensional feature data from them; Performing state mapping processing on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed in different division modes; Calling the trained Markov model to perform probability calculation on the multidimensional feature state sequence to generate corresponding feature state transition probabilities, and performing comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and the feature state transition probability is compared with the feature state transition probability matrix group to obtain corresponding comparison results; outputting a warning signal according to the comparison result; Before performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence, the method further includes: Acquire multiple groups of historical sound signals and extract corresponding multi-dimensional feature data therefrom; Dividing the characteristic value interval of the multidimensional characteristic data according to a preset distance step length to generate a corresponding state mapping table; Combining a plurality of groups of state mapping tables to generate a state mapping table group; Before calling the trained Markov model to perform probability calculation on the multi-dimensional feature state sequence to generate the corresponding feature state transition probability, the method further includes: Acquire multiple groups of normal historical sound signals and extract corresponding multi-dimensional feature data from them; Performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence; Performing Markov hypothesis processing on the multidimensional characteristic state sequence, screening out multidimensional characteristic state sequences that are all Markov processes, generating corresponding characteristic state transition probabilities, and converting them into characteristic state transition probability matrices; A plurality of groups of the characteristic state transition probability matrices are combined to generate a group of characteristic state transition probability matrices.

2. The abnormal sound detection method according to claim 1, It is characterized in that The Markov assumption is that for a random variable at a certain moment, it only depends on the random variable at the previous moment, that is, p(X t |X 0 ,X 1 ,...,X t-1 )=p(X t |X t-1 ), where t = 1, 2, ..., a random sequence that satisfies the Markov property is called a Markov process.

3. The abnormal sound detection method according to claim 2, It is characterized in that The characteristic state transition probability is calculated according to the formula p ij = p(X t =i|X t-1 =j) is converted into the characteristic state transition probability matrix, where i = 1, 2, ...; j = 1, 2, ..., p ij ≥0,∑p ij =1.

4. An abnormal sound detection device, It is characterized in that include: A data acquisition unit, used to acquire sound signals of industrial equipment and extract multi-dimensional feature data therefrom; A data processing unit, used for performing state mapping processing on the multidimensional feature data according to a state mapping table group to generate a multidimensional feature state sequence, wherein the state mapping table group is composed of a plurality of state mapping tables constructed by using different division methods and different implementation functions; A data comparison unit is used to call the trained Markov model to perform probability calculation on the multidimensional feature state sequence, generate corresponding feature state transition probabilities, and perform comparison processing on the feature state transition probabilities to generate corresponding comparison results, wherein the Markov model internally stores a feature state transition probability matrix group composed of a plurality of different feature state transition probability matrices, and compares the feature state transition probability with the feature state transition probability matrix group to obtain corresponding comparison results; An early warning unit, configured to output an early warning signal according to the comparison result; Before performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence, the method further includes: Acquire multiple groups of historical sound signals and extract corresponding multi-dimensional feature data therefrom; Dividing the characteristic value interval of the multidimensional characteristic data according to a preset distance step length to generate a corresponding state mapping table; Combining a plurality of groups of state mapping tables to generate a state mapping table group; Before calling the trained Markov model to perform probability calculation on the multi-dimensional feature state sequence to generate the corresponding feature state transition probability, the method further includes: Acquire multiple groups of normal historical sound signals and extract corresponding multi-dimensional feature data from them; Performing state mapping processing on the multidimensional feature data according to the state mapping table group to generate a multidimensional feature state sequence; Performing Markov hypothesis processing on the multidimensional characteristic state sequence, screening out multidimensional characteristic state sequences that are all Markov processes, generating corresponding characteristic state transition probabilities, and converting them into characteristic state transition probability matrices; A plurality of groups of the characteristic state transition probability matrices are combined to generate a group of characteristic state transition probability matrices.

5. An abnormal sound detection device, It is characterized in that The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, an abnormal sound detection method according to any one of claims 1 to 3 is implemented.

6. A readable storage medium, It is characterized in that A computer program is stored, and the computer program can be executed by a processor of the device where the storage medium is located to implement an abnormal sound detection method as described in any one of claims 1 to 3.

Images