Detection method and device for combustion heater

Abstract

The invention relates to a detection method and device for a combustion heater. The method comprises the steps that monitoring data of target equipment in a preset time period and deformation data of the target equipment are acquired; respectively extracting data features of the monitoring data and the deformation data; combining the data features of the monitoring data and the deformation data through a feature combination technology to obtain target data features; and obtaining feature distribution information of the target data feature, inputting the feature distribution information into a pre-constructed target model, and determining an equipment state of the target equipment. The equipment is comprehensively monitored through various types of sensors, the characteristic distribution of the monitoring data is deeply analyzed by using the data model, and the current state of the target equipment is determined, so that an efficient and accurate diagnosis result is realized, and powerful support is provided for equipment maintenance and fault prevention.

Description

Technical Field

[0001] The present application relates to the field of testing technology, and in particular to a detection method, device, electronic equipment and storage medium for a combustion heater. Background Art

[0002] In the research of mechanical fault diagnosis, more than 70% of the faults are manifested as equipment vibration. The most common, reliable and effective method is to obtain equipment status information by collecting vibration signals containing rich fault information.

[0003] However, in actual operation, when the equipment is in a variable speed condition, the fault characteristic components contained in the vibration signal of the faulty bearing will show time-varying non-stationary characteristics. Fourier transform is a method of extracting signal features from the overall time domain and cannot capture local features. It has significant limitations in analyzing non-stationary signals.

[0004] Therefore, there is an urgent need to develop a detection method and device for combustion heaters to solve one or more of the above-mentioned problems. Summary of the invention

[0005] In view of this, in order to solve the above technical problems or part of the technical problems, the embodiments of the present invention provide a detection method and device for a combustion heater;

[0006] In a first aspect, the present application provides a detection method for a combustion heater, the method comprising:

[0007] Acquire monitoring data of a target device within a preset time period and deformation data of the target device, wherein the monitoring data is data collected by multiple types of sensors installed in the target device;

[0008] respectively extracting data features of the monitoring data and the deformation data;

[0009] Combining the monitoring data and the data features of the deformation data by a feature combination technology to obtain target data features;

[0010] The feature distribution information of the target data feature is obtained, and the feature distribution information is input into a pre-built target model to determine the device state of the target device.

[0011] In a possible implementation manner, respectively extracting data features of the monitoring data and the deformation data includes:

[0012] Performing a first digital conversion on the monitoring data and the deformation data respectively to obtain a first carrier signal and a second carrier signal;

[0013] respectively obtaining carrier scattering coefficients of the first carrier signal and the second carrier signal;

[0014] Based on the carrier scattering coefficients of the first carrier signal and the second carrier signal, respectively determine signal scattering matrix characteristics of the first carrier signal and the second carrier signal;

[0015] Data features of the monitoring data and the deformation data are obtained according to signal scattering matrix features of the first carrier signal and the second carrier signal.

[0016] In a possible implementation manner, the carrier scattering coefficient includes a first carrier scattering coefficient and a second carrier scattering coefficient;

[0017] The obtaining of the carrier scattering coefficients of the first carrier signal and the second carrier signal respectively includes:

[0018] Acquire the signal amplitudes of the first carrier signal and the second carrier signal respectively to obtain a first amplitude and a second amplitude;

[0019] Low-pass filtering is performed on the first amplitude and the second amplitude respectively to obtain first carrier scattering coefficients of the first carrier signal and the second carrier signal;

[0020] The first carrier scattering coefficients of the first carrier signal and the second carrier signal are respectively subjected to first data conversion to obtain second carrier scattering coefficients of the first carrier signal and the second carrier signal.

[0021] In a possible implementation manner, performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain second carrier scattering coefficients of the first carrier signal and the second carrier signal includes:

[0022] Performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain a third carrier signal and a fourth carrier signal;

[0023] The carrier scattering coefficients of the third carrier signal and the fourth carrier signal are respectively acquired to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal.

[0024] In a possible implementation manner, before inputting the target data feature into a pre-built target model to obtain the device state of the target device, the method further includes:

[0025] Get the running status of the target device;

[0026] According to the running status, establishing a running tag;

[0027] The running label is added to the corresponding target data feature.

[0028] In a possible implementation manner, acquiring feature distribution information of the target data feature includes:

[0029] Inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map;

[0030] Compressing the spatial size of the initial feature map to obtain a target feature map;

[0031] Obtain a set of feature vectors corresponding to the target feature map;

[0032] The feature distribution information of the feature vectors in the feature vector set is used as the feature distribution information of the target data feature.

[0033] In a possible implementation, the convolutional neural network includes an initial convolution layer, a defective block, and a splicing layer;

[0034] The step of inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map includes:

[0035] Inputting the target data features into the initial convolution layer of the pre-trained convolutional neural network, and outputting initial feature information;

[0036] Inputting the target data features into the incomplete block of the pre-trained convolutional neural network, and outputting supplementary feature information;

[0037] The initial feature information and the supplementary feature information are input into a pre-trained concatenation layer, and the initial feature information and the supplementary feature information are concatenated to obtain an initial feature map.

[0038] In a possible implementation, the target model construction process includes:

[0039] Obtaining a historical device status record of a target device, wherein the historical device status record includes historical status data and historical operating status;

[0040] Extracting data features of each historical state data in the historical state record;

[0041] Obtaining a training sample set according to the data features of each historical state data in the historical state record and the operating state corresponding to each historical state data;

[0042] The initial model is trained using the training sample set to obtain a target model.

[0043] In a possible implementation, the target model construction process further includes:

[0044] Select test data from the historical device status records of the target device;

[0045] Acquire test feature distribution information of the test data and actual operating status corresponding to the test data;

[0046] Inputting the test feature distribution information into the target model to obtain a test result;

[0047] Using a confusion matrix to obtain the actual operating state corresponding to the test result and the test data, and determining the model parameters of the target model;

[0048] In the case where the model parameters do not conform to the preset rules, the step of training the initial model using the training sample set to obtain the target model is re-executed.

[0049] In a second aspect, the present application provides a detection device for a combustion heater, the device comprising:

[0050] An acquisition module, used to acquire monitoring data of a target device and deformation data of the target device within a preset time period, wherein the monitoring data is data collected by multiple types of sensors installed in the target device;

[0051] An extraction module, used to extract data features of the monitoring data and the deformation data respectively;

[0052] A combination module, used for combining the monitoring data and the data features of the deformation data by using a feature combination technology to obtain a target data feature;

[0053] The determination module is used to obtain the feature distribution information of the target data feature, and input the feature distribution information into a pre-built target model to determine the device state of the target device.

[0054] The above technical solution provided by the embodiment of the present application has the following advantages over the prior art: the method provided by the embodiment of the present application realizes comprehensive monitoring of the equipment through various types of sensors, and then uses neural network technology to effectively extract key features in the monitoring data, capture the frequency characteristics of the signal, and enhance the robustness and accuracy of subsequent equipment monitoring. Then, the data model is used to conduct an in-depth analysis of the characteristic distribution of the monitoring data to determine the current status of the target equipment, thereby achieving efficient and accurate diagnostic results and providing strong support for equipment maintenance and fault prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0057] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0058] Figure 1 A schematic flow chart of a detection method for a combustion heater provided in an embodiment of the present application;

[0059] Figure 2 A schematic flow chart of another detection method for a combustion heater provided in an embodiment of the present application;

[0060] Figure 3 A schematic diagram of a wavelet scattering network framework provided in an embodiment of the present application;

[0061] Figure 4 A schematic diagram of the structure of a defective block provided in an embodiment of the present application;

[0062] Figure 5 A schematic diagram of the structure of a detection device for a combustion heater provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0064] The disclosure below provides many different embodiments or examples to implement different structures of the present invention. In order to simplify the disclosure of the present invention, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present invention. In addition, the present invention can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0065] The main problems with combustion heaters are detonation, unstable combustion, and combustion oscillation; the problems with combustion heaters are mainly manifested in the obvious difference between vibration and strain signals during detonation or unstable combustion. Some combustion heaters are only equipped with vibration sensors. Further analysis of vibration, temperature, and strain signals reveals that they can all show abnormalities in the combustion heater. During equipment maintenance, vibration data is a major performance indicator of the equipment. The strength of vibration is directly related to the operating status of the equipment. Excessive vibration poses a huge threat to the safety of equipment and personnel. Therefore, vibration characteristics are an important indicator of whether the equipment is operating normally. In terms of diagnostic technology, it is relatively easy to collect the vibration signal of the equipment, and the vibration signal includes a rich range of operating conditions of the equipment. The equipment used to monitor vibration abnormalities is also relatively simple. The entire testing and diagnostic process is carried out during the normal operation of the equipment and does not affect actual production activities.

[0066] Taking the health monitoring and fault diagnosis of rolling bearings, a key mechanical component, as an example, the fault characteristic components in the vibration signal of the faulty bearing under variable speed conditions show time-varying non-stationarity, while Fourier transform is a signal analysis method that extracts features from the global time domain and cannot extract local features. It has great limitations for the analysis of non-stationary signals: the short-time Fourier transform method is easy to implement, but the resolution is single; the time-frequency analysis method represented by wavelet transform is a signal analysis method based on the time and frequency reference system, which has been proven to be applicable to the description of non-stationary signals. Since wavelet transform has the characteristics of local deformation stability and multi-scale, it can effectively extract the local features of the signal, but because it changes with time, it is easy to miss the signal features, and it is not suitable for the analysis of time-varying non-stationary signals and the construction of feature vectors.

[0067] In order to solve the problem in the prior art that when the equipment is in a variable speed condition, the fault characteristic components contained in the vibration signal of the faulty bearing will show time-varying non-stationary characteristics, and the traditional Fourier transform method cannot effectively capture these local characteristics, resulting in obvious deficiencies in analyzing non-stationary signals, the embodiment of the present application provides a detection method and device for a combustion heater, which utilizes the powerful capabilities of a convolutional neural network (CNN) to extract features from multiple sensor data on the equipment, and also introduces a wavelet transform network to further extract features, and uses these features as input data for the convolutional neural network. Through this method of combining a wavelet transform network and a convolutional neural network, the efficiency and accuracy of fault diagnosis are significantly improved, thereby achieving the purpose of reducing detection costs and reducing human resource consumption.

[0068] Figure 1 A flow chart of a detection method for a combustion heater provided in an embodiment of the present application is shown as follows: Figure 1 As shown, the method specifically includes:

[0069] S101, acquiring monitoring data of a target device within a preset time period and deformation data of the target device, wherein the monitoring data is data collected by multiple types of sensors installed in the target device;

[0070] In this embodiment, vibration sensors, strain sensors, and temperature sensors are installed at key positions of the combustion heater to monitor the vibration and strain conditions in the heater and the combustion temperature respectively; deformation sensors are installed on the inner wall of the combustion chamber to monitor the deformation data of the heater; and operating data of key positions are collected in real time through multiple types of sensors, and the data is transmitted to a data processing unit, so that the data processing unit uses advanced signal processing algorithms to analyze and process the collected vibration, strain and temperature signals to identify potential failure modes.

[0071] S102, extracting data features of the monitoring data and the deformation data respectively;

[0072] Data feature extraction methods can enable models such as deep learning to better learn the patterns in the data, thereby improving prediction accuracy, classification precision, etc.

[0073] In this embodiment, in order to improve the accuracy of the detection of the combustion heater, the data features of the monitoring data and the deformation data are extracted respectively, and then the data features are input into the target model to facilitate in-depth analysis and processing of the monitoring data.

[0074] S103, combining the monitoring data and the data features of the deformation data by a feature combination technology to obtain target data features;

[0075] Feature combination technology combines multiple original data features into new features to mine more complex relationships in the data. In this way, information that cannot be reflected by a single feature can be discovered, thereby improving the performance and predictive ability of the model.

[0076] In this embodiment, after obtaining the data features of monitoring data and deformation data, data fusion technology is used to comprehensively analyze the data of different sensors to improve the accuracy and reliability of fault detection. Through data feature combination technology, information from different sensors can be effectively integrated to obtain more comprehensive and accurate equipment status information than a single sensor.

[0077] S104: Acquire feature distribution information of the target data feature, and input the feature distribution information into a pre-built target model to determine the device state of the target device.

[0078] In this embodiment, by obtaining the feature distribution information of the target data features, the distribution of the data in different dimensions is revealed, such as frequency, probability, and possible outliers, and then the feature distribution information is input into a pre-built target model. The target model is designed based on previous equipment data training and can analyze and judge the current state of the target device according to the input feature distribution information. Finally, the target model is used to effectively determine whether the target device is in a normal working state, or whether there are potential faults and performance degradation problems, thereby improving fault diagnosis efficiency and diagnostic accuracy, thereby saving detection costs and manpower.

[0079] The detection method for combustion heaters provided in the embodiment of the present application realizes comprehensive monitoring of the equipment through various types of sensors, and can cover all aspects of the equipment operation; then, by using neural network technology, it can effectively extract key features in the monitoring data, capture the frequency characteristics of the signal, and thus enhance the robustness and accuracy of subsequent equipment monitoring; then, through the data model, the characteristic distribution of the monitoring data is deeply analyzed to determine the current state of the target equipment, thereby achieving efficient and accurate diagnosis results, and providing strong support for equipment maintenance and fault prevention; and in the process of key feature extraction, the ResNet model is used to receive the characteristic coefficients extracted by the wavelet scattering network, and further abstracted and classified through the deep learning method. The ResNet model solves the gradient vanishing problem in the deep network through the residual block structure, better handles high-dimensional complex data, and improves the performance and stability of the model, thereby achieving efficient and accurate diagnosis results in the fault diagnosis of the combustion heater.

[0080] The structure of the wavelet scattering convolutional network is similar to that of CNN, except that its filter is a predetermined wavelet filter, and the parameters of the wavelet filter do not need to be learned through training samples, and its network is non-feedback. The signal is transformed by calculating the semi-discrete wavelet transform, and then through the nonlinear modulus operation, the obtained signal feature expression has excellent characteristics such as translation invariance and deformation stability, which just meets the basic requirements of feature extractors in machine learning. In addition, these basic properties have been rigorously theoretically proved in mathematics, which makes up for the lack of theoretical support for CNN. Therefore, it is proposed to introduce the wavelet scattering network into the design of fault diagnosis algorithm.

[0081] Wavelet transform is a signal processing technique that decomposes a signal into wavelet components of different scales (frequencies) and positions. Wavelet is a function with finite duration and oscillation characteristics, and the signal is analyzed by translating and scaling the wavelet. Compared with Fourier transform, which mainly focuses on the frequency component of the signal, wavelet transform considers both frequency and time (or space) information, and can more effectively analyze non-stationary signals.

[0082] Taking the health monitoring and fault diagnosis of rolling bearings, a key mechanical component, as an example, the fault characteristic components in the vibration signal of the faulty bearing under variable speed conditions show time-varying non-stationarity, while Fourier transform is a signal analysis method that extracts features from the global time domain and cannot extract local features. It has great limitations for the analysis of non-stationary signals: the short-time Fourier transform method is easy to implement, but the resolution is single; the time-frequency analysis method represented by wavelet transform is a signal analysis method based on the time and frequency reference system, which has been proven to be applicable to the description of non-stationary signals. Since wavelet transform has the characteristics of local deformation stability and multi-scale, it can effectively extract the local features of the signal, but because it changes with time, it is easy to miss the signal features, and it is not suitable for the analysis of time-varying non-stationary signals and the construction of feature vectors.

[0083] Figure 2 A flow chart of another detection method for a combustion heater provided in an embodiment of the present application is shown as follows: Figure 2 As shown, the method specifically includes:

[0084] S201, performing a first digital conversion on the monitoring data and the deformation data respectively to obtain a first carrier signal and a second carrier signal;

[0085] In this embodiment, wavelet transform is performed on the monitored data and the deformed data respectively to obtain wavelet signals of the first layer, which are used as the first carrier signal and the second carrier signal respectively.

[0086] Furthermore, after acquiring the first carrier signal, a modulus of the first carrier signal is calculated;

[0087] Exemplarily, wavelet transform is applied to the vibration data v(t) to obtain the first carrier signal W of the first layer. v1 :

[0088]

[0089] Calculate the modulus of the first carrier signal |W v1 |:

[0090]

[0091] S202, respectively obtaining carrier scattering coefficients of the first carrier signal and the second carrier signal;

[0092] In this embodiment, the acquired first carrier signal and the second carrier signal are respectively subjected to low-pass filtering to filter out high-frequency noise, so as to obtain corresponding carrier scattering coefficients;

[0093] Exemplarily, the result of the modulus operation is low-pass filtered to obtain the carrier signal transmission coefficient S of the first layer. v1 :

[0094]

[0095] S203, determining signal scattering matrix characteristics of the first carrier signal and the second carrier signal respectively based on the carrier scattering coefficients of the first carrier signal and the second carrier signal;

[0096] In the present invention, the acquired carrier scattering coefficients of the first carrier signal and the second carrier signal are calculated and analyzed respectively to determine the signal scattering matrix characteristics of the first carrier signal and the second carrier signal respectively.

[0097] For example, for S v1 Perform wavelet transform, modular operation and nonlinear averaging again to obtain the carrier signal transmission coefficient S of the second layer v2 :

[0098]

[0099] Similarly, the strain data s1(t) and s2(t) are transformed to obtain S s12 ,S s22 .

[0100] S204: Obtain data characteristics of the monitoring data and the deformation data according to signal scattering matrix characteristics of the first carrier signal and the second carrier signal.

[0101] In this embodiment, by comparing and analyzing the scattering matrix characteristics of the first carrier signal and the second carrier signal, the specific state or failure mode of the device is identified, and the detection parameters can be adjusted in real time according to the changes in the scattering matrix characteristics to adapt to different detection environments and conditions, thereby ensuring the accuracy and reliability of the detection results.

[0102] The detection method for combustion heaters provided in the embodiment of the present application can effectively extract stable characteristic coefficients with multi-scale characteristics from the original sensor data by adopting wavelet scattering network technology, which not only captures the changes of the signal in the time dimension, but also accurately reflects the characteristics of the signal in the frequency dimension. It can significantly improve the robustness of subsequent models when facing various complex situations, and also enhance the accuracy of the model when performing prediction and classification tasks.

[0103] The scattering coefficients of all orders are aggregated to describe the wavelet scattering characteristics of the input THz detection signal, where is the maximum decomposition order.

[0104] In an optional solution of the embodiment of the present invention, the carrier scattering coefficient includes a first carrier scattering coefficient and a second carrier scattering coefficient;

[0105] The obtaining of the carrier scattering coefficients of the first carrier signal and the second carrier signal respectively includes:

[0106] Acquire the signal amplitudes of the first carrier signal and the second carrier signal respectively to obtain a first amplitude and a second amplitude;

[0107] Low-pass filtering is performed on the first amplitude and the second amplitude respectively to obtain first carrier scattering coefficients of the first carrier signal and the second carrier signal;

[0108] The first carrier scattering coefficients of the first carrier signal and the second carrier signal are respectively subjected to first data conversion to obtain second carrier scattering coefficients of the first carrier signal and the second carrier signal.

[0109] In this embodiment, the signal amplitudes of the first carrier signal and the second carrier signal, i.e., the first amplitude and the second amplitude, are measured and acquired respectively, and then low-pass filtering operations are performed on the two amplitude parameters respectively to extract low-frequency components in the first carrier signal and the second carrier signal, thereby obtaining the first carrier scattering coefficients of the first carrier signal and the second carrier signal; finally, the first carrier scattering coefficients of the first carrier signal and the second carrier signal obtained above are further subjected to data conversion processing to obtain more accurate carrier scattering parameters, i.e., the second carrier scattering coefficients of the first carrier signal and the second carrier signal; in this way, the first carrier scattering coefficients and the second carrier scattering coefficients are comprehensively analyzed to achieve accurate interpretation of the detection signal, effectively enhance the anti-interference capability of the detection signal, and improve the accuracy of signal processing.

[0110] In an optional solution of the embodiment of the present invention, performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal includes:

[0111] Performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain a third carrier signal and a fourth carrier signal;

[0112] The carrier scattering coefficients of the third carrier signal and the fourth carrier signal are respectively acquired to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal.

[0113] In this embodiment, the first acquired carrier signal and the second carrier signal are first processed, and the two signals are respectively subjected to the first data conversion to obtain two new carrier signals, namely the third carrier signal and the fourth carrier signal, and then the carrier scattering coefficients of the two newly acquired carrier signals, namely the third carrier signal and the fourth carrier signal, are respectively obtained, namely the second carrier scattering coefficients of the first carrier signal and the second carrier signal. This method can realize the accurate interpretation of the detection signal, and because the carrier scattering coefficients of the third carrier signal and the fourth carrier signal are calculated by a specific algorithm, the anti-interference ability of the detection signal can be effectively enhanced, and the accuracy of signal processing can be improved.

[0114] like Figure 3 As shown in Figure 1, the wavelet scattering network is implemented through deep convolution, which iterates the traditional wavelet transform, nonlinear and average operators. Let f(t) be the signal to be analyzed, and use the low-pass filter φ and the wavelet function ψ to construct a filter covering all the frequencies contained in the signal. φ(t) is the low-pass filter that provides the local translation invariant feature of f on the scale T, and Λ k Indicates frequency resolution, multi-scale high-pass filter Constructed by wavelet function ψ, the scattering coefficient is S k Convolution S0f(t)=f*Φ j (t) generates local translation invariant features at scale T, but it also leads to the loss of high-frequency information. These lost high-frequency information W1 can be restored by wavelet modulus transform, which can be expressed as

[0115]

[0116] The first-order scattering coefficient is given by φ j The average wavelet modulus coefficient is obtained

[0117]

[0118] Recovering the lost information by averaging, S1f(t) can be regarded as the low-frequency component, and the complementary high-frequency coefficient W2 can be extracted by the following formula, which can be expressed as

[0119]

[0120] The second-order scattering coefficient S2 is further defined and expressed as

[0121]

[0122] Iterating the above process defines the wavelet modular convolution, which is expressed as

[0123]

[0124] U mf(t) and φ j Taking the average, the m-order scattering coefficient is

[0125]

[0126] The final scattering matrix feature is expressed as

[0127] Sf(t)={S m f(t)}0≤m≤l;

[0128] In an optional solution of the embodiment of the present invention, before inputting the target data feature into a pre-built target model to obtain the device state of the target device, the method further includes:

[0129] Get the running status of the target device;

[0130] According to the running status, establishing a running tag;

[0131] The running label is added to the corresponding target data feature.

[0132] In this embodiment, the current operating status information of the target device is first obtained; then, corresponding operating tags are created based on the obtained device operating status information, and these created operating tags are added and associated with the corresponding target data features to ensure that the target data features closely correspond to the actual operating status of the device, thereby improving the accuracy of device status prediction.

[0133] In an optional solution of the embodiment of the present invention, the acquiring feature distribution information of the target data feature includes:

[0134] Inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map;

[0135] Compressing the spatial size of the initial feature map to obtain a target feature map;

[0136] Obtain a set of feature vectors corresponding to the target feature map;

[0137] The feature distribution information of the feature vectors in the feature vector set is used as the feature distribution information of the target data feature.

[0138] In this embodiment, the target data features to be analyzed are first input into a pre-trained convolutional neural network model, and a set of initial feature maps are obtained through processing of this model. The initial feature maps are then compressed in spatial dimensions to obtain more concise and concentrated target feature maps. Further, a corresponding feature vector set is extracted from the target feature map. Finally, the feature distribution information of each feature vector in the feature vector set is combined as the feature distribution information of the final target data features, which can effectively represent the complexity and diversity of the target data features, and can more accurately identify and predict the operating status of the equipment, thereby providing strong data support for the maintenance and management of the equipment. In addition, by using a pre-trained convolutional neural network model and utilizing existing knowledge and experience, the efficiency and accuracy of feature extraction can be improved, the reliance on a large amount of labeled data can be reduced, and the implementation cost can be reduced.

[0139] like Figure 4 As shown in the figure, the residual structure adds a branch line, combines the input with the result of the convolution operation on the input, and then uses it as the input of the next layer through the activation function. This branch line allows the gradient from the deep layer to pass unimpeded to the previous layer, so that the shallow network parameters can be effectively trained. DenseNet is even more extreme, connecting each layer with each other through branches to achieve cross-layer connection. The gradient obtained by each layer is the addition of the gradients of the previous layers.

[0140] Assume that the network consists of L residual blocks, x l represents the input of the lth residual lock, F(x l ,W l ) represents the residual part of the lth residual block, W l Represents the corresponding weight, then the output of l residual blocks is expressed as follows:

[0141]

[0142] The output of the Lth layer can be expressed as follows:

[0143]

[0144] From the above formula, we can see that no matter which layer in the residual block, there is the original input x1, which ensures that the gradient will not disappear during back propagation. In the back propagation stage, the chain derivation formula shows that due to the existence of the residual structure, no matter how deep the network is, there is 1 in the chain derivation result. The result of this formula is not always -1, thus avoiding the gradient disappearance.

[0145] The chain derivation formula is as follows:

[0146]

[0147] In an optional solution of the embodiment of the present invention, the convolutional neural network includes an initial convolution layer, a defective block, and a splicing layer;

[0148] The step of inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map includes:

[0149] Inputting the target data features into the initial convolution layer of the pre-trained convolutional neural network, and outputting initial feature information;

[0150] Inputting the target data features into the incomplete block of the pre-trained convolutional neural network, and outputting supplementary feature information;

[0151] The initial feature information and the supplementary feature information are input into a pre-trained concatenation layer, and the initial feature information and the supplementary feature information are concatenated to obtain an initial feature map.

[0152] The incomplete block convolutional neural network is a neural network architecture specially designed to process data containing incomplete blocks (such as images, audio, etc.). It is an improvement on the traditional convolutional neural network (CNN) to better cope with incomplete data. In this network, the design of convolutional layers, pooling layers, and fully connected layers will take into account how to effectively extract useful information from incomplete blocks and how to mitigate the negative impact of incomplete blocks on tasks such as feature extraction and classification.

[0153] In this embodiment, the feature information of the target data is first input into an initial convolutional layer of a pre-trained convolutional neural network. Through the processing of this initial convolutional layer, we can obtain a set of initial feature information; then, the same target data features are input into the incomplete block of the same pre-trained convolutional neural network. The incomplete block will further process the input features according to its internal structure, thereby outputting a set of supplementary feature information; finally, the initial feature information obtained above and the supplementary feature information are input into a pre-trained splicing layer together. The splicing layer will perform an effective splicing operation on the two sets of feature information, and finally obtain an initial feature map.

[0154] Exemplarily, the extracted feature vector X and the corresponding label Y (e.g., normal combustion, incomplete combustion, etc.) are combined into a training data set. A ResNet model is constructed. The ResNet model is characterized by using residual blocks to construct the network, which improves the expression ability and performance of the model, thereby avoiding gradient vanishing and gradient explosion during training and accelerating network convergence.

[0155] The feature vector X is input into the ResNet model and goes through the following processing steps:

[0156] Step 1: Initial convolutional layer, including convolution, batch normalization, ReLU activation, and maximum pooling.

[0157] Step 2: The residual block includes multiple convolutional layers, batch normalization, ReLU activation, and residual connection.

[0158] Step 3: The layer consisting of multiple residual blocks involves gradually reducing the spatial size of the feature map.

[0159] Step 4: Global average pooling involves compressing the spatial size of the feature map to 1.

[0160] Step 5: The fully connected layer includes flattening the feature vector, performing classification, and outputting the probability distribution.

[0161] In an optional solution of the embodiment of the present invention, the process of constructing the target model includes:

[0162] Obtaining a historical device status record of a target device, wherein the historical device status record includes historical status data and historical operating status;

[0163] Extracting data features of each historical state data in the historical state record;

[0164] Obtaining a training sample set according to the data features of each historical state data in the historical state record and the operating state corresponding to each historical state data;

[0165] The initial model is trained using the training sample set to obtain a target model.

[0166] In this embodiment, in order to establish a target model, the historical device status record of the target device is first obtained, the historical status data and the historical operating status information are obtained, and then the data features of each historical status data are extracted from the historical status record, which are used to analyze and understand the key to the device status. After the data features are extracted, a training sample set is constructed based on the data features and the operating status corresponding to each historical status data. This set will serve as the basis for subsequent model training. Finally, the initial model is deeply trained using the training sample set to obtain a more accurate and reliable target model. This method takes into account the real-time operating status and historical data of the equipment, can effectively predict possible problems with the equipment, and give maintenance suggestions in a timely manner; in addition, the model parameters can be adjusted according to the real-time data to adapt to changes in the equipment status, thereby improving the accuracy and efficiency of detection.

[0167] In an optional solution of the embodiment of the present invention, the process of constructing the target model further includes:

[0168] Select test data from the historical device status records of the target device;

[0169] Acquire test feature distribution information of the test data and actual operating status corresponding to the test data;

[0170] Inputting the test feature distribution information into the target model to obtain a test result;

[0171] Using a confusion matrix to obtain the actual operating state corresponding to the test result and the test data, and determining the model parameters of the target model;

[0172] In the case where the model parameters do not conform to the preset rules, the step of training the initial model using the training sample set to obtain the target model is re-executed.

[0173] In this embodiment, during the process of building the target model, the test feature distribution information is also used to test whether the target model meets the model requirements.

[0174] Specifically, firstly, representative test data is selected from the historical device status records of the target device, the test feature distribution information of the test data is obtained, and the actual operating status corresponding to the test data is recorded at the same time, and then the test feature distribution information is input into the target model, and the corresponding test results are obtained through the calculation and processing of the model; further, the test results and the actual operating status corresponding to the test data are compared and analyzed using a confusion matrix to determine whether the model parameters of the target model are accurate; if it is found that the model parameters do not meet the preset rules and standards, then it is necessary to re-execute the step of training the initial model using the training sample set to obtain the target model, so as to ensure that a target model that meets the requirements is obtained.

[0175] Exemplarily, the test data is input into the trained ResNet model, and the diagnosis result is output. The confusion matrix is ​​used to count the number of actual and predicted results. This paper belongs to a binary classification problem. Both the predicted results and the actual situation have two situations: normal and fault. After combining them, four situations will appear, as shown in Table 1. Normal and fault are represented by 0 and 1 respectively.

[0176] Table 1

[0177]

[0178]

[0179] The meanings of TP, FP, FN, and TN are: TP is the number of faults that are consistent with the actual situation; FP is the number of faults that are normal in actual situation; FN is the number of normals that are faults in actual situation; TN is the number of normals that are consistent with actual situation. Therefore, we get:

[0180] Model sensitivity = TP / (TP+FN) = 67%

[0181] Model specificity = TN / (FP+TN) = 91%

[0182] Model accuracy (ACC) = (TP + TN) / (TP + FP + FN + TN) = 90.6%

[0183] According to the above results, it can be found that the diagnostic results of this method are relatively excellent, and the model conforms to the preset rules.

[0184] Figure 5 A schematic diagram of a detection device for a combustion heater provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the device specifically includes:

[0185] An acquisition module 501 is used to acquire monitoring data of a target device and deformation data of the target device within a preset time period, wherein the monitoring data is data collected by multiple types of sensors installed in the target device;

[0186] An extraction module 502, used to extract data features of the monitoring data and the deformation data respectively;

[0187] A combining module 503 is used to combine the data features of the monitoring data and the deformation data to obtain target data features through feature combination technology;

[0188] The determination module 504 is used to obtain the feature distribution information of the target data feature, and input the feature distribution information into a pre-built target model to determine the device state of the target device.

[0189] In one possible embodiment, the extraction module 502 is also used to perform a first digital conversion on the monitoring data and the deformation data, respectively, to obtain a first carrier signal and a second carrier signal; respectively obtain carrier scattering coefficients of the first carrier signal and the second carrier signal; respectively determine signal scattering matrix characteristics of the first carrier signal and the second carrier signal based on the carrier scattering coefficients of the first carrier signal and the second carrier signal; and obtain data characteristics of the monitoring data and the deformation data based on the signal scattering matrix characteristics of the first carrier signal and the second carrier signal.

[0190] In one possible embodiment, the extraction module 502 is also used to perform a first digital conversion on the monitoring data and the deformation data, respectively, to obtain a first carrier signal and a second carrier signal; respectively obtain carrier scattering coefficients of the first carrier signal and the second carrier signal; respectively determine signal scattering matrix characteristics of the first carrier signal and the second carrier signal based on the carrier scattering coefficients of the first carrier signal and the second carrier signal; and obtain data characteristics of the monitoring data and the deformation data based on the signal scattering matrix characteristics of the first carrier signal and the second carrier signal.

[0191] In a possible implementation, the extraction module 502 is further used to perform a first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal, respectively, to obtain a third carrier signal and a fourth carrier signal; and to obtain the carrier scattering coefficients of the third carrier signal and the fourth carrier signal, respectively, to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal.

[0192] In a possible implementation, the extraction module 502 is further configured to obtain an operating status of the target device; establish an operating tag according to the operating status; and add the operating tag to a corresponding target data feature.

[0193] In one possible implementation, the determination module 504 is also used to input the target data features into a pre-trained convolutional neural network to obtain an initial feature map; compress the spatial size of the initial feature map to obtain a target feature map; obtain a set of feature vectors corresponding to the target feature map; and use the feature distribution information of the feature vectors in the feature vector set as the feature distribution information of the target data features.

[0194] In one possible implementation, the determination module 504 is further used to input the target data features into an initial convolutional layer of a pre-trained convolutional neural network, and output initial feature information; input the target data features into a defective block of a pre-trained convolutional neural network, and output supplementary feature information; input the initial feature information and the supplementary feature information into a pre-trained splicing layer, and splice the initial feature information and the supplementary feature information to obtain an initial feature map.

[0195] In one possible implementation, the device further includes a construction module 505, which is used to obtain a historical device status record of the target device, wherein the historical device status record includes historical status data and historical operating status; extract data features of each historical status data in the historical status record; obtain a training sample set based on the data features of each historical status data in the historical status record and the operating status corresponding to each historical status data; and use the training sample set to train the initial model to obtain a target model.

[0196] In one possible implementation, the construction module 505 is also used to select test data from the historical device status record of the target device; obtain test feature distribution information of the test data, and the actual operating status corresponding to the test data; input the test feature distribution information into the target model to obtain a test result; use a confusion matrix to obtain the test result and the actual operating status corresponding to the test data, and determine the model parameters of the target model; if the model parameters do not meet the preset rules, re-execute the step of training the initial model using the training sample set to obtain the target model.

[0197] The detection device for the combustion heater provided in this embodiment can be as follows Figure 5 The detection device for the combustion heater shown in FIG. can be performed as follows: Figure 1-4 All the steps in the detection of combustion heaters are implemented Figure 1-4 The technical effect of the detection of combustion heaters shown is shown in Figure 1-4 For the sake of brevity, the relevant description is not repeated here.

[0198] The device embodiments described above are merely illustrative, wherein the device components described as separate components may or may not be physically separated, and the components displayed as device components may or may not be physical device components, that is, they may be located in one place, or may be distributed on multiple network device components. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0199] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0200] It should be understood that the terms used in the text are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used in the text may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described in the text are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0201] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A detection method for a combustion heater, characterized in that: include: Acquire monitoring data of a target device within a preset time period and deformation data of the target device, wherein the monitoring data is data collected by multiple types of sensors installed in the target device; respectively extracting data features of the monitoring data and the deformation data; Combining the monitoring data and the data features of the deformation data by a feature combination technology to obtain target data features; The feature distribution information of the target data feature is obtained, and the feature distribution information is input into a pre-built target model to determine the device state of the target device.

2. The method according to claim 1, characterized in that The extracting the data features of the monitoring data and the deformation data respectively includes: Performing a first digital conversion on the monitoring data and the deformation data respectively to obtain a first carrier signal and a second carrier signal; respectively obtaining carrier scattering coefficients of the first carrier signal and the second carrier signal; Based on the carrier scattering coefficients of the first carrier signal and the second carrier signal, respectively determine signal scattering matrix characteristics of the first carrier signal and the second carrier signal; Data features of the monitoring data and the deformation data are obtained according to signal scattering matrix features of the first carrier signal and the second carrier signal.

3. The method according to claim 2, characterized in that The carrier scattering coefficient includes a first carrier scattering coefficient and a second carrier scattering coefficient; The obtaining of the carrier scattering coefficients of the first carrier signal and the second carrier signal respectively includes: Acquire the signal amplitudes of the first carrier signal and the second carrier signal respectively to obtain a first amplitude and a second amplitude; Low-pass filtering is performed on the first amplitude and the second amplitude respectively to obtain first carrier scattering coefficients of the first carrier signal and the second carrier signal; The first carrier scattering coefficients of the first carrier signal and the second carrier signal are respectively subjected to first data conversion to obtain second carrier scattering coefficients of the first carrier signal and the second carrier signal.

4. The method according to claim 3, characterized in that: The performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal includes: Performing first data conversion on the first carrier scattering coefficients of the first carrier signal and the second carrier signal respectively to obtain a third carrier signal and a fourth carrier signal; The carrier scattering coefficients of the third carrier signal and the fourth carrier signal are respectively acquired to obtain the second carrier scattering coefficients of the first carrier signal and the second carrier signal.

5. The method according to claim 3, characterized in that: Before inputting the target data feature into a pre-built target model to obtain the device state of the target device, the method further includes: Get the running status of the target device; According to the running status, establishing a running tag; The running label is added to the corresponding target data feature.

6. The method according to claim 1, characterized in that The acquiring of the characteristic distribution information of the target data characteristic comprises: Inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map; Compressing the spatial size of the initial feature map to obtain a target feature map; Obtain a set of feature vectors corresponding to the target feature map; The feature distribution information of the feature vectors in the feature vector set is used as the feature distribution information of the target data feature.

7. The method according to claim 6, characterized in that The convolutional neural network includes an initial convolution layer, a defective block, and a splicing layer; The step of inputting the target data features into a pre-trained convolutional neural network to obtain an initial feature map includes: Inputting the target data features into the initial convolution layer of the pre-trained convolutional neural network, and outputting initial feature information; Inputting the target data features into the incomplete block of the pre-trained convolutional neural network, and outputting supplementary feature information; The initial feature information and the supplementary feature information are input into a pre-trained concatenation layer, and the initial feature information and the supplementary feature information are concatenated to obtain an initial feature map.

8. The method according to claim 5, characterized in that The target model construction process includes: Obtaining a historical device status record of a target device, wherein the historical device status record includes historical status data and historical operating status; Extracting data features of each historical state data in the historical state record; Obtaining a training sample set according to the data features of each historical state data in the historical state record and the operating state corresponding to each historical state data; The initial model is trained using the training sample set to obtain a target model.

9. The method according to claim 8, characterized in that The target model construction process also includes: Select test data from the historical device status records of the target device; Acquire test feature distribution information of the test data and actual operating status corresponding to the test data; Inputting the test feature distribution information into the target model to obtain a test result; Using a confusion matrix to obtain the actual operating state corresponding to the test result and the test data, and determining the model parameters of the target model; In the case where the model parameters do not conform to the preset rules, the step of training the initial model using the training sample set to obtain the target model is re-executed.

10. A detection device for a combustion heater, characterized in that: include: An acquisition module, used to acquire monitoring data of a target device and deformation data of the target device within a preset time period, wherein the monitoring data is data collected by multiple types of sensors installed in the target device; An extraction module, used to extract data features of the monitoring data and the deformation data respectively; A combination module, used for combining the monitoring data and the data features of the deformation data by using a feature combination technology to obtain a target data feature; The determination module is used to obtain the feature distribution information of the target data feature, and input the feature distribution information into a pre-built target model to determine the device state of the target device.

Images