Equipment start-stop state classification model training method, equipment start-stop state classification method, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an equipment start-stop state classification model training method, an equipment start-stop state classification method, electronic equipment and a storage medium, and relates to the technical field of industrial monitoring. According to the method, multiple pieces of sample data of the mechanical equipment are obtained, the sample data are collected according to a preset frequency, each piece of sample data comprises a magnetic flux waveform, a vibration waveform and a real start-stop state of the mechanical equipment at the same moment, the sample data are preprocessed, multiple sample sets are obtained, and the sample sets comprise multiple pieces of sample data with continuous collection moments; and inputting the sample sets into the equipment start-stop state classification model in sequence for training until the equipment start-stop state classification model reaches a preset condition. Therefore, the start-stop state of the mechanical equipment can be monitored in real time by using the trained equipment start-stop state classification model.

Description

Technical Field

[0001] The present invention relates to the field of industrial monitoring technology, and in particular to a method for training a device start / stop state classification model, a method for classifying device start / stop states, an electronic device and a storage medium. Background Art

[0002] With the development of industrial automation and intelligence, modern industrial production is increasingly dependent on mechanical equipment. Mechanical equipment plays a key role in the production process, and its start and stop status directly affects production efficiency, product quality and production safety. Production safety is an important aspect of concern for enterprises. Abnormal start and stop of mechanical equipment often indicates potential failure risks and may even cause safety accidents. For example, high-speed rotating mechanical parts may cause serious damage to equipment and personnel when they suddenly stop. Real-time monitoring of the start and stop status of equipment can detect abnormal conditions in time, take preventive maintenance measures, and avoid safety accidents.

[0003] In the prior art, start-stop status monitoring is often performed through manual periodic inspection and instrument monitoring. However, manual periodic inspection requires workers to regularly patrol and inspect the equipment to determine the operating status of the equipment. Due to the limited frequency of manual monitoring, abnormal start-stop situations of the equipment are often not discovered in time. In addition, the experience and judgment ability of workers will also affect the accuracy of monitoring, which can easily lead to misjudgment or missed judgment. Instrument monitoring still requires manual reading and recording of data, making it difficult to achieve real-time monitoring. Summary of the invention

[0004] In view of this, an object of an embodiment of the present invention is to provide a device start / stop state classification model training method, a device start / stop state classification method, an electronic device and a storage medium to at least partially improve the above-mentioned problem.

[0005] In order to achieve the above purpose, the technical solution adopted by the embodiment of the present invention is as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for training a device start / stop state classification model, the method comprising:

[0007] Acquire multiple sample data of the mechanical equipment; the sample data are collected at a preset frequency, and each sample data includes a magnetic flux waveform, a vibration waveform and a real start / stop state of the mechanical equipment at the same time;

[0008] Preprocessing each of the sample data to obtain a plurality of sample sets; the sample sets include a plurality of the sample data collected consecutively at the time of collection;

[0009] Each of the sample sets is input into the equipment start-stop state classification model in turn for training until the equipment start-stop state classification model meets preset conditions.

[0010] Optionally, the preprocessing of each sample data to obtain multiple sample sets includes:

[0011] For each of the sample data, Fourier transform is performed on the magnetic flux waveform and the vibration waveform to obtain corresponding magnetic flux spectrum and vibration spectrum, and the magnetic flux spectrum and the vibration spectrum are added to the sample data;

[0012] For each of the sample data, a dimensionless feature of the sample data is calculated according to the magnetic flux waveform, the magnetic flux spectrum, the vibration waveform and the vibration spectrum, and the dimensionless feature is added to the sample data;

[0013] Each of the sample data is divided into a plurality of incompletely overlapping sample sets.

[0014] Optionally, the device start / stop state classification model includes a convolutional neural network and a classification network, and the step of inputting each of the sample sets into the device start / stop state classification model for training includes:

[0015] Inputting each of the magnetic flux waveforms, each of the magnetic flux spectra, each of the vibration waveforms, and each of the vibration spectra in the sample set into the convolutional neural network to obtain corresponding magnetic flux waveform features, magnetic flux spectrum features, vibration waveform features, and vibration spectrum features;

[0016] Input each of the magnetic flux waveform features, each of the magnetic flux spectrum features, each of the vibration waveform features, each of the vibration spectrum features and each of the dimensionless features into the classification network, and output a start-stop state classification result;

[0017] A loss value is calculated according to the start-stop state classification result and the actual start-stop state of the last sample data in the sample set, and the start-stop state classification model is iteratively updated through back propagation according to the loss value.

[0018] Optionally, the classification network includes a multi-layer perceptron, a long short-term memory network and a fully connected classification layer, and each of the magnetic flux waveform features, each of the magnetic flux spectrum features, each of the vibration waveform features, each of the vibration spectrum features and each of the dimensionless features is input into the classification network, and a start-stop state classification result is output, including:

[0019] Inputting the magnetic flux waveform feature, magnetic flux spectrum feature, vibration waveform feature and vibration spectrum feature corresponding to the last sample data of the sample set into the multi-layer perceptron for separate feature extraction to obtain a single-time feature representation;

[0020] Inputting the single-moment feature representation into the fully connected classification layer to obtain the single-moment start-stop classification probability;

[0021] Performing feature fusion on each of the magnetic flux waveform features, each of the magnetic flux spectrum features, each of the vibration waveform features, each of the vibration spectrum features, and each of the dimensionless features in the sample set to obtain a multi-modal and multi-dimensional time series feature;

[0022] Inputting the multimodal and multidimensional time series features into the long short-term memory network to perform time series feature processing to obtain a time series feature representation;

[0023] Input the time series feature representation into the fully connected classification layer to obtain the time series start-stop classification probability;

[0024] The single-moment start-stop classification probability and the time sequence start-stop classification probability are weightedly fused to obtain a start-stop state classification result.

[0025] Optionally, weighted fusion of the single-moment start-stop classification probability and the sequential start-stop classification probability to obtain a start-stop state classification result includes:

[0026] The final classification probability is calculated through the weighted fusion formula;

[0027] When the final classification probability is greater than a preset value, the start-stop state classification result is a stop state;

[0028] When the final classification probability is less than or equal to a preset value, the start-stop state classification result is the start state.

[0029] Optionally, the weighted fusion formula is:

[0030] P=w 1 P m +w 2 P s

[0031] Among them, P is the final classification probability, P m is the single-moment start-stop classification probability, P s is the time sequence start-stop classification probability, w 1 is the weight of the start-stop classification probability at a single moment, w 2 is the weight of the time series start-stop classification probability;

[0032] The preset value is greater than 0.5.

[0033] Optionally, the method further comprises:

[0034] Using the Adam optimizer to update the parameters of the equipment start / stop state classification model;

[0035] Introducing a batch normalization layer and a Dropout layer into the device start / stop state classification model;

[0036] Distributed training and parallel computing technology are used for the equipment start-stop state classification model, and each of the sample sets and the equipment start-stop state classification model is distributed to multiple computing nodes for training and computing;

[0037] The device start / stop state classification model is compressed by using a parameter pruning method, and the parameters are converted from high-precision floating-point numbers to low-precision integers.

[0038] In a second aspect, an embodiment of the present invention provides a method for classifying the start and stop states of a device, the method comprising:

[0039] Obtain the magnetic flux waveform and vibration waveform of the mechanical equipment at the current moment and multiple consecutive moments before the current moment collected by the vibration and magnetic flux integrated sensor;

[0040] Input each of the magnetic flux waveforms and each of the vibration waveforms into a device start / stop state classification model to obtain the start / stop state of the mechanical device at the current moment; the device start / stop state classification model is trained by the method described in the first aspect above.

[0041] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of the first aspect above is implemented; and / or the method described in the second aspect above is implemented.

[0042] In a fourth aspect, an embodiment of the present invention provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any one of the first aspect above is implemented; and / or the method described in the second aspect above is implemented.

[0043] The embodiments of the present invention provide a method for training a model for classifying the start-stop state of an equipment, a method for classifying the start-stop state of an equipment, an electronic device, and a storage medium. The vibration signal is combined with the working condition magnetic flux signal. Through the complementary ability between different signals and the redundant information provided by multidimensional data, the model is made to perform more stably under different working conditions, and the stability and reliability of the equipment start-stop state monitoring model can be improved. The vibration waveform and the working condition magnetic flux waveform are used to generate a spectrum through FFT, and then combined with the extracted dimensionless features. Through multimodal data fusion, the characteristics of the signal can be more comprehensively captured, which helps to reduce the underfitting phenomenon of the model and improve the generalization and learning ability. Based on the continuity of the start-stop state of the equipment, the single-moment feature representation and the time series feature representation are fused to obtain the final result of the start-stop classification, so that the final start-stop result has a stronger generalization ability. Thus, the start-stop state of the mechanical equipment can be better monitored.

[0044] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 A schematic structural block diagram of an electronic device provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of a flow chart of a method for training a device start / stop state classification model provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a flow chart of step S220 provided in an embodiment of the present invention;

[0049] Figure 4 A schematic structural block diagram of a device start-stop state classification model provided by an embodiment of the present invention;

[0050] Figure 5 A schematic diagram of a flow chart of step S230 provided in an embodiment of the present invention;

[0051] Figure 6 Another schematic structural block diagram of a device start / stop state classification model provided by an embodiment of the present invention;

[0052] Figure 7 A schematic diagram of a flow chart of step S232 provided in an embodiment of the present invention;

[0053] Figure 8 A flowchart of a method for classifying the start and stop states of equipment provided by an embodiment of the present invention.

[0054] Icons: 100-electronic device; 101-memory; 102-communication interface; 103-processor; 104-bus; 300-equipment start / stop state classification model; 310-convolutional neural network; 320-classification network; 321-multilayer perceptron; 322-long short-term memory network; 323-fully connected classification layer. DETAILED DESCRIPTION

[0055] 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, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents 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 creative work are within the scope of protection of the present invention.

[0057] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0058] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0059] As described in the background technology, with the development of industrial automation and intelligence, modern industrial production is increasingly dependent on mechanical equipment. Mechanical equipment plays a key role in the production process, and its start and stop status directly affects production efficiency, product quality and production safety. Production safety is an important aspect of concern to enterprises. Abnormal start and stop of mechanical equipment often indicates potential failure risks and may even cause safety accidents. For example, high-speed rotating mechanical parts may cause serious damage to equipment and personnel when they suddenly stop. Real-time monitoring of the start and stop status of equipment can detect abnormal situations in time, take preventive maintenance measures, and avoid the occurrence of safety accidents.

[0060] In the prior art, the start-stop status is often monitored through manual periodic inspection and instrument monitoring. Manual periodic inspection means that workers patrol and inspect the equipment at predetermined time intervals to determine its operating status. This method usually includes visual inspection, auditory inspection, and tactile inspection. The operating parameters of the equipment are monitored through various instruments installed on the equipment (such as pressure gauges, thermometers, ammeters, etc.). Workers regularly read the data of these instruments and determine the start-stop status of the equipment based on preset thresholds.

[0061] However, due to the limited frequency of manual monitoring, abnormal start and stop conditions of equipment are often not discovered in time. In addition, workers' experience and judgment ability will also affect the accuracy of monitoring, which can easily lead to misjudgment or missed judgment. Instrument monitoring still requires manual reading and recording of data, making it difficult to achieve real-time monitoring. At the same time, the accuracy and reliability of the instrument itself may be affected by environmental factors.

[0062] Based on the above situation, an embodiment of the present invention provides a device start-stop status classification model training method, a device start-stop status classification method, an electronic device and a storage medium. By acquiring multiple sample data of mechanical equipment, each sample data is preprocessed to obtain multiple sample sets, and the sample set includes multiple sample data continuous at the collection time. Each sample set is input into the device start-stop status classification model for training in turn until the equipment start-stop status classification model meets the preset conditions. The trained equipment start-stop status classification model is used to monitor the start-stop status of the mechanical equipment in real time.

[0063] To implement the process steps and functions of each example of the present invention, please refer to Figure 1 , Figure 1 A schematic structural block diagram of an electronic device provided by an embodiment of the present invention. The electronic device 100 includes a memory 101 and a processor 103, and the memory 101 and the processor 103 are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses 104 or signal lines. The memory 101 can be used to store software programs and modules, and the processor 103 executes the software programs and modules stored in the memory 101 to perform various functional applications and data processing.

[0064] The electronic device 100 may be, but is not limited to, a personal computer (PC), a server, a distributed computer, etc. It is understandable that the electronic device 100 is not limited to a physical server, but may also be a virtual machine on a physical server, a virtual machine built on a cloud platform, etc., which can provide the same functions as the server or virtual machine. The operating system of the electronic device 100 may be, but is not limited to, a Windows system, a Linux system, etc.

[0065] Among them, the memory 101 can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.

[0066] The communication connection between the electronic device 100 and an external device is achieved through at least one communication interface 102 (which can be wired or wireless).

[0067] The processor 103 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the embodiment of the present invention can be completed by the hardware integrated logic circuit in the processor 103 or the instructions in the form of software. The processor 103 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0068] Understandably, Figure 1 The structure shown is for illustration only. The electronic device 100 may also include Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown. Figure 1 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0069] The following is an exemplary description of the equipment start-stop state classification model training method provided by the present invention. Specifically, Figure 2 A flow chart of a device start-stop state classification model training method provided by an embodiment of the present invention, see Figure 2 , the execution subject of this method can be the above Figure 1 The electronic device 100 shown in FIG. 1 includes: Figure 2 The following steps are shown:

[0070] S210: Acquire multiple sample data of the mechanical equipment; wherein the sample data is collected at a preset frequency, and each sample data includes a magnetic flux waveform, a vibration waveform, and a real start / stop state of the mechanical equipment at the same moment.

[0071] The sample data is collected during the daily operation of mechanical equipment. Vibration and magnetic flux integrated sensors can be installed at multiple key parts of mechanical equipment (such as motor shafts, transmission components, etc.). The vibration and magnetic flux integrated sensors can collect vibration signals and magnetic flux signals at the same time. Vibration and magnetic flux integrated sensors are installed based on multiple measuring points to ensure that signals based on vibration and magnetic flux can be collected comprehensively and accurately. At the same time, a unified data collection frequency and duration can be defined, that is, the frequency of collecting vibration signals and magnetic flux signals is the same, to ensure that subtle changes in equipment operation can be captured. Optionally, the range of the collection frequency can be set to 2-2000Hz.

[0072] For example, a signal is collected once every 2 hours, and the signal data collected this time is used as a sample data, including a flux waveform and a vibration waveform, and the start and stop status of the mechanical equipment at that time is recorded to obtain a complete sample data.

[0073] Since a single signal cannot fully describe the state of mechanical equipment, combining the vibration signal with the flux signal can provide more comprehensive feature information. At the same time, the complementary ability between different signals can enhance the discrimination ability of subsequent model construction.

[0074] S220: Preprocess each sample data to obtain multiple sample sets; wherein the sample set includes multiple sample data collected at consecutive times.

[0075] The data collected by the integrated vibration and magnetic flux sensor often have problems such as noise, missing values ​​and outliers. Data cleaning and noise reduction technology can be used to normalize the sample data, unify the vibration waveform and magnetic flux waveform to the same length, and process the missing values ​​of the data to remove the missing data in the data set.

[0076] All sample data are time-series and can be divided into multiple samples. Each sample data in a sample set is collected continuously. For example, the sample data acquired in sequence are 1, 2, 3, ..., 100, and each sample set includes 5 sample data. Then a sample set can be 1, 2, 3, 4, 5, or 2, 3, 4, 5, 6, ..., or 96, 97, 98, 99, 100. Of course, the number of sample data in a sample set can be different, which will result in multiple sample sets.

[0077] Optionally, the processed sample set is divided into a data set. In order to fully verify the effectiveness of the model, while ensuring that there are enough samples, the sample set is divided into a training set, a validation set, and a test set according to the principle of making the test set ratio as large as possible, with a total distribution ratio of 7:1:2.

[0078] S230: Inputting each sample set into the equipment start / stop state classification model in sequence for training until the equipment start / stop state classification model meets preset conditions.

[0079] The divided sample sets are input into the equipment start-stop status classification model for training in turn. The training of the overall model can be to obtain the error between the network output and the expected target through forward propagation, judge whether the network has converged, and through back propagation and weight modification, use the BP back propagation algorithm to propagate the obtained error back to each node layer by layer, and update the weights. Repeat the forward propagation and back propagation iterative calculation process until the network converges. The model's operating effect is preliminarily evaluated through the verification set to verify the model's generalization ability. After multiple experiments and multiple iterations, the model loss value is gradually close to zero, so that the model is in a convergence state, the training is completed, and a mature equipment start-stop status classification model is obtained.

[0080] The method divides the sample data into multiple sample sets including multiple sample data with continuous collection times, inputs the sample sets into the equipment start-stop state classification model for training, obtains a mature input equipment start-stop state classification model for training, and then uses the equipment start-stop state classification model to monitor the start-stop state of the mechanical equipment in real time.

[0081] In one possible implementation, in order to more fully capture the characteristics of the signal and reduce the underfitting phenomenon of the model, the flux waveform and the vibration waveform can be further processed, see Figure 3 , Figure 3 A schematic flow chart of step S220 provided in an embodiment of the present invention, wherein step S220 may include the following steps:

[0082] S221: For each sample data, perform Fourier transform on the magnetic flux waveform and the vibration waveform respectively to obtain the corresponding magnetic flux spectrum and vibration spectrum respectively, and add the magnetic flux spectrum and the vibration spectrum to the sample data.

[0083] The flux waveform and vibration waveform only provide information in the time domain. The flux waveform and vibration waveform can be Fourier transformed to generate flux spectrum and vibration spectrum. The shape data can provide information in the time domain, and the spectrum data can provide information in the frequency domain. Combining these two types of data can more comprehensively capture the characteristics of the signal, thereby improving the performance of the model.

[0084] At this time, the single sample data includes the flux waveform, vibration waveform, flux spectrum and vibration spectrum.

[0085] S222: For each sample data, a dimensionless feature of the sample data is calculated according to the magnetic flux waveform, the magnetic flux spectrum, the vibration waveform and the vibration spectrum, and the dimensionless feature is added to the sample data.

[0086] Furthermore, the dimensionless features of the sample data can be calculated as the input of the model. By adding the dimensionless features to the deep learning model, additional information can be provided, thereby improving the prediction performance and accuracy of the model.

[0087] Dimensionless characteristics may include: kurtosis factor, skewness factor, waveform factor, margin factor, pulse factor, and peak factor.

[0088] S223: Divide each sample data into a plurality of incompletely overlapping sample sets.

[0089] The processed sample data is divided into multiple incompletely overlapping sample sets according to the above method.

[0090] In one possible implementation, see Figure 4 , Figure 4 A schematic structural block diagram of a device start-stop state classification model provided in an embodiment of the present invention, wherein the device start-stop state classification model 300 may include a convolutional neural network 310 and a classification network 320, see Figure 5 , Figure 5 A flowchart of step S230 provided in an embodiment of the present invention, the step of inputting each sample set into the device start / stop state classification model for training may include:

[0091] S231: Input each magnetic flux waveform, each magnetic flux spectrum, each vibration waveform and each vibration spectrum in the sample set into the convolutional neural network to obtain the corresponding magnetic flux waveform characteristics, magnetic flux spectrum characteristics, vibration waveform characteristics and vibration spectrum characteristics.

[0092] First, feature extraction is performed on each flux waveform, each flux spectrum, each vibration waveform, and each vibration spectrum in the sample set. Using convolutional neural networks, four one-dimensional convolutional neural networks can be used to extract features from vibration waveforms and spectra, flux waveforms and spectra. Since the length of the original collected waveform is generally large, 16K in this example, that is, 16K features, a larger convolution kernel is set in the first layer, such as kernel_size=16, to obtain a larger receptive field, so that each convolution kernel in the subsequent layers can learn as long as possible in the original one-dimensional data, that is, it can learn as many features in the data as possible. At the same time, deep separable convolution is used instead of traditional convolution operations to reduce the number of parameters.

[0093] S232: Input each magnetic flux waveform feature, each magnetic flux spectrum feature, each vibration waveform feature, each vibration spectrum feature and each dimensionless feature into a classification network, and output a start-stop state classification result.

[0094] S233: Calculate the loss value based on the start-stop state classification result and the actual start-stop state of the last sample data in the sample set, and iteratively update the start-stop state classification model through back propagation based on the loss value.

[0095] Among them, the loss function can use cross entropy loss.

[0096] Optionally, the Figure 6 , Figure 6 Another schematic structural block diagram of a device start-stop state classification model provided by an embodiment of the present invention, the classification network 320 may include a multi-layer perceptron 321, a long short-term memory network 322 and a fully connected classification layer 323, see Figure 7 , Figure 7 A schematic flow chart of step S232 provided in an embodiment of the present invention, step S232 may include the following steps:

[0097] S2321: Input the magnetic flux waveform feature, magnetic flux spectrum feature, vibration waveform feature and vibration spectrum feature corresponding to the last sample data of the sample set into a multi-layer perceptron for individual feature extraction to obtain a single-time feature representation.

[0098] For example, the sample data in the sample set are 2, 3, 4, 5, and 6, and the last sample data is 6. The flux waveform features, flux spectrum features, vibration waveform features, and vibration spectrum features in the sample data are extracted separately through a multi-layer perceptron to obtain a single-time feature representation.

[0099] S2322: Input the single-moment feature representation into the fully connected classification layer to obtain the single-moment start-stop classification probability.

[0100] S2323: Fusing the magnetic flux waveform features, magnetic flux spectrum features, vibration waveform features, vibration spectrum features and dimensionless features in the sample set to obtain multi-modal and multi-dimensional time series features.

[0101] Each feature in the sample set is temporally continuous. Each feature can be combined with the dimensionless feature and weightedly fused using the attention mechanism to obtain a new fused feature. The new fused feature is multimodal, multidimensional and temporal. The fused feature includes multidimensional data of vibration waveform and flux waveform, as well as multimodal data of vibration spectrum, flux spectrum and dimensionless feature, and each sample data is temporally continuous.

[0102] S2324: Input the multimodal and multidimensional time series features into the long short-term memory network for time series feature processing to obtain a time series feature representation.

[0103] Using long short-term memory networks to process multimodal and multi-dimensional time series features can detect long-term dependencies in time series and fuse information from different moments, which helps to perform effective data fusion at the feature level.

[0104] S2325: Input the time series feature representation into the fully connected classification layer to obtain the time series start-stop classification probability.

[0105] The fully connected classification layer may include a fully connected layer and a Softmax function. The fully connected layer is used to map the learned features to the final category. Each neuron usually corresponds to a category and outputs the probability of each category through the Softmax function.

[0106] S2326: Perform weighted fusion of the single moment start-stop classification probability and the time series start-stop classification probability to obtain a start-stop state classification result.

[0107] Optionally, step S2326 may include:

[0108] The final classification probability is calculated through the weighted fusion formula.

[0109] The weighted fusion formula can be:

[0110] P=w 1 P m +w 2 P s

[0111] Among them, P is the final classification probability, P m is the single-moment start-stop classification probability, P s is the time sequence start-stop classification probability, w 1 is the weight of the start-stop classification probability at a single moment, w 2 is the weight of the temporal start-stop classification probability.

[0112] For example, the probability of single-moment start-stop classification is 0.5, the probability of sequential start-stop classification is 0.6, and w 1 is 0.7, w 2 is 0.3, then the final classification probability is 0.5*0.7+0.6*0.3=0.53. 1 、w 2 It can be adjusted automatically during model training.

[0113] When the final classification probability is greater than the preset value, the start-stop state classification result is the stop state.

[0114] When the final classification probability is less than or equal to the preset value, the start-stop state classification result is the start state.

[0115] The preset value is the posterior probability cutpoint. Generally, the preset value is 0.5. If the final classification probability is greater than 0.5, it is in the shutdown state. If the final classification probability is less than or equal to 0.5, it is in the startup state.

[0116] In real industrial scenarios, the role of equipment start and stop judgment is to provide a basis for subsequent equipment operation plans, health status monitoring, etc. If the equipment start state is judged as a shutdown state in actual production, it means that the equipment is in a shutdown state. If the equipment fails at this time, it may lead to more serious consequences or even accidents. Therefore, in practice, it is necessary to minimize the misjudgment of start and stop as shutdown. Although judging shutdown as start will cause certain interference to the industrial site, it will not cause serious economic losses. Therefore, in this embodiment, the preset value is greater than 0.5, for example, it can be set to 0.6. In the above example, the final classification probability is 0.53. In conventional design, the mechanical equipment is judged to be in a shutdown state, but in this embodiment, it is in a start state.

[0117] Furthermore, in order to reduce the number of parameters of the overall model deployed in the vibration and magnetic flux integrated sensor, the method may also include:

[0118] Use the Adam optimizer to update the parameters of the equipment start / stop status classification model.

[0119] Compared with other optimizers, the Adam optimizer has the fastest convergence speed and high accuracy. It is also easier to adjust parameters and can set a larger initial learning rate for adaptive learning. The learning rate has a certain range in each iteration, making the parameters relatively stable.

[0120] Batch normalization layer and Dropout layer are introduced into the device start / stop status classification model.

[0121] The batch normalization layer calculates the sliding mean, sliding standard deviation, gamma, and beta through multiple iterations and back propagation, thereby converting the output data to the same distribution to solve the gradient explosion and gradient vanishing problems, accelerate the convergence of the model, and improve the generalization ability of the model. In each iteration, the Dropout layer stops some neurons from working at a certain ratio, which can effectively reduce the occurrence of overfitting and plays a role of regularization to a certain extent. At the same time, it also eliminates the joint adaptability between neurons and enhances the generalization ability of the model.

[0122] Distributed training and parallel computing technology are used for the equipment start-stop status classification model, and each sample set and the equipment start-stop status classification model are distributed to multiple computing nodes for training and computing.

[0123] Due to the large scale of training sample data, in order to speed up the training of the model, distributed training and parallel computing technology can be used. Data and models are distributed to multiple computing nodes, and training and computing are performed simultaneously. Through parallel processing and data parallelism, training efficiency is improved and training time is shortened.

[0124] The parameter pruning method is used to compress the equipment start / stop status classification model, and the parameters are converted from high-precision floating-point numbers to low-precision integers.

[0125] In order to meet the needs of lower-level computer deployment and further reduce the storage and computing requirements of the model, the parameter pruning method is used to compress the model. At the same time, the parameters are converted from high-precision floating-point numbers to low-precision integers to reduce the storage space and computing power of the model.

[0126] Furthermore, the embodiment of the present invention also provides a method for classifying the start and stop states of a device, see Figure 8 , Figure 8 A schematic diagram of a flow chart of a method for classifying the start and stop states of a device provided in an embodiment of the present invention, the method comprising:

[0127] S410: Obtain the magnetic flux waveform and vibration waveform of the mechanical equipment at the current moment and multiple consecutive moments before the current moment collected by the vibration and magnetic flux integrated sensor.

[0128] S420: Input each magnetic flux waveform and each vibration waveform into the equipment start-stop state classification model to obtain the start-stop state of the mechanical equipment at the current moment; wherein the equipment start-stop state classification model is trained by the above-mentioned equipment start-stop state classification model training method.

[0129] Optionally, based on the comprehensive consideration of the service life and power consumption of the vibration and flux integrated sensor, when deployed at the edge segment, since the vibration and flux integrated sensor is greatly affected by the battery power consumption, this embodiment embeds the convolutional neural network into the vibration and flux integrated sensor, and the subsequent convolutional neural network runs at the acquisition station end. This approach has many advantages. First, it can reduce the number of parameters of the overall model deployed in the vibration and flux integrated sensor as much as possible, so that the service life of the vibration and flux integrated sensor is increased as much as possible; second, the feature vector is much smaller than the original data, which reduces the amount of data transmission when transmitting to the upper computer, and reduces the bandwidth demand and transmission cost; third, the CNN has a large amount of calculation. Deploying it at the sensor end can use special hardware acceleration (such as ASIC or FPGA) to improve efficiency, while the subsequent module parameters are large, but the amount of calculation is small, and it can run efficiently on the general hardware of the acquisition station, avoiding the waste of computing resources at the sensor end. Fourth, the feature extraction layer of the convolutional neural network is closer to the representation of the data itself. The subsequent upgrade and iteration of the model only requires adjusting the output processing module at the acquisition station end, which greatly increases the sustainable iteration of the model. Therefore, this architecture combines the advantages of edge computing and deep learning to achieve an intelligent system with high efficiency, low latency and high reliability.

[0130] In this embodiment, the sample size of the model training is 40,000 devices with one year of data. The devices mainly include pumps, fans, compressors, etc., with a total of about 30 million waveform data. The final model is obtained through distributed integrated training (without adjusting the preset Cutpoint value). At the same time, another 8,000 devices are used as a test set to test the final effect of the model. The evaluation results are shown in Table 1:

[0131] Table 1

[0132]

[0133] As can be seen from Table 1, based on actual industrial data, for the binary classification problem of start-stop status recognition, the model trained by the embodiment of the present invention is significantly better than the effect of judging the start and stop of equipment through thresholds and rules on the basis of the same data set. What is more noteworthy is that the number of devices with an accuracy rate less than 0.6 in a single device is only 82 in the model provided by the present invention, which is far less than the 334 based on the threshold and rule judgment method. This shows that in real industrial scenarios, the model proposed by the embodiment of the present invention can provide more accurate start and stop status for as many devices as possible. Furthermore, the model adjusted by the preset value has a higher accuracy rate. This model adjusts the classification probability of the model output through the cutpoint, which can more effectively reduce the false alarm rate of the system and is more in line with real industrial scenarios.

[0134] In summary, an embodiment of the present invention provides a method for training a device start-stop state classification model, a method for classifying a device start-stop state, an electronic device, and a storage medium, which combine a vibration signal with an operating condition magnetic flux signal, and through the complementary ability between different signals and the redundant information provided by multidimensional data, the model is more stable under different working conditions, and the stability and reliability of the equipment start-stop state monitoring model can be improved; the vibration waveform and the operating condition magnetic flux waveform are generated by FFT to generate a spectrum, and then combined with the extracted dimensionless features, through multimodal data fusion, the characteristics of the signal can be more comprehensively captured, which helps to reduce the underfitting phenomenon of the model and improve the generalization and learning ability; based on the continuity of the start-stop state of the equipment, the single-moment feature representation and the time series feature representation are fused to obtain the final result of the start-stop classification, so that the final start-stop result has a stronger generalization ability; the classification probability of the model output is adjusted by cutpoint, which can more effectively reduce the false alarm rate of the system and be more in line with the actual industrial scenario.

[0135] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a module, a program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0136] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0137] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a computer-readable storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0138] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0139] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for training a device start-stop status classification model, characterized in that: The method comprises: Acquire multiple sample data of the mechanical equipment; the sample data are collected at a preset frequency, and each sample data includes a magnetic flux waveform, a vibration waveform and a real start / stop state of the mechanical equipment at the same time; Preprocessing each of the sample data to obtain a plurality of sample sets; the sample sets include a plurality of the sample data collected consecutively at the time of collection; Each of the sample sets is input into the equipment start-stop state classification model in turn for training until the equipment start-stop state classification model meets preset conditions.

2. The method according to claim 1, characterized in that The preprocessing of each sample data to obtain multiple sample sets includes: For each of the sample data, Fourier transform is performed on the magnetic flux waveform and the vibration waveform to obtain corresponding magnetic flux spectrum and vibration spectrum, and the magnetic flux spectrum and the vibration spectrum are added to the sample data; For each of the sample data, a dimensionless feature of the sample data is calculated according to the magnetic flux waveform, the magnetic flux spectrum, the vibration waveform and the vibration spectrum, and the dimensionless feature is added to the sample data; Each of the sample data is divided into a plurality of incompletely overlapping sample sets.

3. The method according to claim 2, characterized in that The equipment start / stop state classification model includes a convolutional neural network and a classification network, and inputting each of the sample sets into the equipment start / stop state classification model for training includes: Inputting each of the magnetic flux waveforms, each of the magnetic flux spectra, each of the vibration waveforms, and each of the vibration spectra in the sample set into the convolutional neural network to obtain corresponding magnetic flux waveform features, magnetic flux spectrum features, vibration waveform features, and vibration spectrum features; Input each of the magnetic flux waveform features, each of the magnetic flux spectrum features, each of the vibration waveform features, each of the vibration spectrum features and each of the dimensionless features into the classification network, and output a start-stop state classification result; A loss value is calculated according to the start-stop state classification result and the actual start-stop state of the last sample data in the sample set, and the start-stop state classification model is iteratively updated through back propagation according to the loss value.

4. The method according to claim 3, characterized in that The classification network includes a multi-layer perceptron, a long short-term memory network and a fully connected classification layer. The magnetic flux waveform features, the magnetic flux spectrum features, the vibration waveform features, the vibration spectrum features and the dimensionless features are input into the classification network to output the start-stop state classification result, including: Inputting the magnetic flux waveform feature, magnetic flux spectrum feature, vibration waveform feature and vibration spectrum feature corresponding to the last sample data of the sample set into the multi-layer perceptron for separate feature extraction to obtain a single-time feature representation; Inputting the single-moment feature representation into the fully connected classification layer to obtain the single-moment start-stop classification probability; Performing feature fusion on each of the magnetic flux waveform features, each of the magnetic flux spectrum features, each of the vibration waveform features, each of the vibration spectrum features, and each of the dimensionless features in the sample set to obtain a multi-modal and multi-dimensional time series feature; Inputting the multimodal and multidimensional time series features into the long short-term memory network to perform time series feature processing to obtain a time series feature representation; Input the time series feature representation into the fully connected classification layer to obtain the time series start-stop classification probability; The single-moment start-stop classification probability and the time sequence start-stop classification probability are weightedly fused to obtain a start-stop state classification result.

5. The method according to claim 4, characterized in that The weighted fusion of the single moment start-stop classification probability and the time sequence start-stop classification probability to obtain the start-stop state classification result includes: The final classification probability is calculated through the weighted fusion formula; When the final classification probability is greater than a preset value, the start-stop state classification result is a stop state; When the final classification probability is less than or equal to a preset value, the start-stop state classification result is the start state.

6. The method according to claim 5, characterized in that The weighted fusion formula is: P=w1P m +w2P s Among them, P is the final classification probability, P m is the single-moment start-stop classification probability, P s is the time sequence start-stop classification probability, w1 is the weight of the single moment start-stop classification probability, and w2 is the weight of the time sequence start-stop classification probability; The preset value is greater than 0.

5.

7. The method according to claim 1, characterized in that The method further comprises: Using the Adam optimizer to update the parameters of the equipment start / stop state classification model; Introducing a batch normalization layer and a Dropout layer into the device start / stop state classification model; Distributed training and parallel computing technology are used for the equipment start-stop state classification model, and each of the sample sets and the equipment start-stop state classification model is distributed to multiple computing nodes for training and computing; The device start / stop state classification model is compressed by using a parameter pruning method, and the parameters are converted from high-precision floating-point numbers to low-precision integers.

8. A method for classifying the start and stop status of equipment, characterized in that: The method comprises: Obtain the magnetic flux waveform and vibration waveform of the mechanical equipment at the current moment and multiple consecutive moments before the current moment collected by the vibration and magnetic flux integrated sensor; Input each of the flux waveforms and each of the vibration waveforms into a device start-stop state classification model to obtain the start-stop state of the mechanical device at the current moment; the device start-stop state classification model is trained by the method described in any one of claims 1 to 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 and / or claim 8 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 and / or claim 8 is implemented.

Images