Equipment characteristic parameter data acquisition and processing method and system based on multi-source heterogeneous data fusion

Abstract

The invention relates to an equipment characteristic parameter data acquisition and processing method and system based on multi-source heterogeneous data fusion. The method comprises the following steps: S1, acquiring an acquisition signal; s2, preprocessing the acquired signals, and performing feature extraction and feature fusion on the preprocessed acquired signals according to data types of the preprocessed acquired signals to generate multi-source feature vectors; s3, performing fault diagnosis or performance evaluation based on different analysis requirements according to the multi-source feature vectors; s4, training the basic feature extraction network according to the increase of the data types of the collected signals, generating a new feature extraction network, and returning to S2; and S5, training the basic performance parameter network according to the increase of the analysis requirements, generating a new performance parameter network, and returning to S3. According to the method, the analysis precision of the characteristic parameter data acquisition and processing system of the multi-distance heterogeneous data equipment can be improved, and the use requirement of the system on the newly-added performance evaluation capability can be met.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion. Background Technology

[0002] Equipment characteristic assessment is crucial for ensuring the safe and efficient operation of equipment systems. As requirements for the reliability and safety of equipment systems increase, there are also higher demands for the reliability and functional diversity of characteristic assessments, such as fault alarms, fault risk warnings, and performance analysis. Simultaneously, an increasing number of different types of sensors are being installed on equipment. This increase in the variety and number of sensors provides richer monitoring data for equipment characteristic assessment; however, in actual monitoring, the data structures, formats, and frequencies acquired from different sensors vary significantly. This results in a large amount of multi-source heterogeneous data generated in single-vehicle and multi-vehicle collaborative systems within multi-modal equipment acquisition systems. This data includes one-dimensional time-series signals and two-dimensional video signals from different sensors, with different storage formats, sampling rates, sampling methods, and dimensions. Given the current situation of multiple sensors, diverse functional requirements, and continuously emerging new data types and functional requirements, it is particularly important to integrate multi-source heterogeneous data and achieve multi-functional characteristic assessment. Summary of the Invention

[0003] To address the technical problems existing in the prior art, the present invention aims to provide a method and system for acquiring and processing equipment characteristic parameters based on multi-source heterogeneous data fusion, thereby improving the system's analytical accuracy and meeting its requirements for evaluating new performance.

[0004] To achieve the above-mentioned objectives, this invention provides a method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion, comprising the following steps:

[0005] Step S1: Acquire the signal;

[0006] Step S2: Preprocess the acquired signal, and according to the data type of the preprocessed acquired signal, use the corresponding feature extraction network to extract features from the preprocessed acquired signal, and fuse the extracted features to generate a multi-source feature vector.

[0007] Step S3: Based on the multi-source feature vectors, and according to different analysis requirements, use the corresponding performance parameter networks to perform fault diagnosis or performance evaluation.

[0008] Step S4: Based on the increase in the data types of the acquired signals, train the basic feature extraction network to generate a new feature extraction network, and return to step S2;

[0009] Step S5: Based on the increased analysis requirements, train the basic performance parameter network to generate a new performance parameter network, and return to step S3.

[0010] According to one technical solution of the present invention, the preprocessed acquired signal is divided into one-dimensional signal data and two-dimensional signal data. The types of one-dimensional signal data include resistance data, current data, AC voltage data, DC voltage data, level signal data, and audio signal data; the types of two-dimensional signal data include image data.

[0011] The feature extraction network for one-dimensional signal data is a one-dimensional convolutional neural network, and the feature extraction network for two-dimensional signal data is a two-dimensional convolutional neural network.

[0012] The performance parameter network is a fully connected layer, consisting of an input layer, several hidden layers, and an output layer.

[0013] According to a technical solution of the present invention, in step S2, the acquired signal is preprocessed, including:

[0014] Step S21: Perform a data integrity check on the acquired signals, including handling missing items and detecting and handling outliers;

[0015] Step S22: Standardize the data format of the acquired signals. The standardization of data format includes standardization of time format, standardization of unit format, and conversion of data type.

[0016] Step S23: Denoise the acquired signals, including sensor data denoising and video data denoising;

[0017] Step S24: Determine the target interception time range based on the maximum sampling period of the acquired signal, and intercept the acquired signal data according to the target interception time range; the intercepted one-dimensional signal data is the continuous signal within the target interception time range, and the intercepted two-dimensional signal data is the last image within the target interception time range.

[0018] Step S25: Unify the data dimensions of the acquired signals.

[0019] According to a technical solution of the present invention, in step S22, the data type conversion includes:

[0020] According to the analysis requirements, the collected text data is converted into numerical data;

[0021] Based on the analysis requirements, the collected numerical data will be converted into categorical data.

[0022] The acquired video signal data is decomposed into one-dimensional audio signals and two-dimensional image sequences.

[0023] According to a technical solution of the present invention, in step S25, the data dimension unification of the one-dimensional signal includes: using the number of sampling points of the signal with the highest sampling frequency within the target intercept time range as the size standard, and interpolating the acquired signals with a lower sampling frequency to the same size.

[0024] Unifying the data dimensions of two-dimensional signals includes: using the image with the highest spatial resolution as the size standard, and filling the background of signals with lower spatial resolutions to make the gray values ​​of the original image's coordinate system origin and eight neighboring regions the same.

[0025] According to one technical solution of the present invention, in step S2, feature fusion is performed by vector concatenation.

[0026] According to one aspect of the present invention, a system for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion is provided, for implementing the above-mentioned method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion, comprising:

[0027] The data acquisition unit is used to classify sensor signals according to the types of sensor signals acquired, and to acquire sensor signals according to the categories of sensor signals.

[0028] A data fusion processing unit is used to preprocess the acquired sensor signals and fuse multi-source heterogeneous data; the data fusion processing unit includes a preprocessing unit and a feature extraction module, the feature extraction module including several feature extraction networks;

[0029] The functional implementation unit is used to perform fault diagnosis and performance evaluation based on the fusion results of multi-source heterogeneous data according to functional requirements; the functional implementation unit includes several performance parameter networks;

[0030] Data storage unit, used to store the results of fault diagnosis and performance evaluation;

[0031] The information display unit is used to display the results of fault diagnosis and performance evaluation.

[0032] According to one technical solution of the present invention, the data acquisition unit includes a core processing module and a power management module, and further includes:

[0033] An ADC acquisition module, connected to the core processing module, is used to convert the acquired analog signals into digital signals and output them to the core processing module.

[0034] An AUTBUS bus processing module, connected to the core processing module, is used to communicate with the data fusion processing unit;

[0035] The clock module is used to provide the reference clock required for the operation of the core processing module and the AUTBUS bus processing module.

[0036] According to one technical solution of the present invention, the core processing module includes:

[0037] The ZYNQ processor is connected to the ADC acquisition module and is used to receive the digital signal output by the ADC acquisition module; the ZYNQ processor is also connected to the AUTBUS bus processing module.

[0038] eMMC storage is used to store received digital signals;

[0039] Ethernet and CAN interfaces are used for communication with the host computer.

[0040] A timing interface is provided for connection to the clock module.

[0041] The DDR3 and QSPI interfaces are used to provide data cache and boot program storage for the ZYNQ processor.

[0042] According to one technical solution of the present invention, the AUTBUS bus processing module includes: a front-end transceiver, an AUTBUS controller, and a core processing module communication interface connected in sequence; the core processing module communication interface adopts an RGMII interface.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The present invention classifies the sensors mounted on the equipment according to the power supply method and signal output type, and modularizes the feature extraction network so that the same type of sensor can use the same feature extraction network, thereby reducing the complexity of the design of the data acquisition unit.

[0045] (2) This invention utilizes multiple performance parameter networks to simultaneously realize the functions of fault and abnormal alarm, risk warning, and performance analysis, avoiding the singleness of traditional methods that only perform multi-source heterogeneous fusion for specific functions.

[0046] (3) The present invention has the ability to keep the original network parameters unchanged when adding new functions or sensor data, which can avoid increasing the time cost of training neural networks when there are new requirements for the feature analysis system. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0048] Figure 1 This diagram illustrates the working principle of an equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion according to an embodiment of the present invention.

[0049] Figure 2 A schematic diagram illustrating the classification of signals acquired according to an embodiment of the present invention;

[0050] Figure 3 This diagram illustrates the architecture of an equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion, as provided in an embodiment of the present invention.

[0051] Figure 4 This schematic diagram illustrates the structure of an equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion provided in an embodiment of the present invention.

[0052] Figure 5 This diagram schematically illustrates the composition of a power management circuit according to an embodiment of the present invention.

[0053] Figure 6 This schematic diagram illustrates the design block diagram of the ADC acquisition module according to an embodiment of the present invention.

[0054] Figure 7 This schematic diagram illustrates the design block diagram of the AUTBUS bus processing module according to an embodiment of the present invention.

[0055] Figure 8 This schematic diagram illustrates the structure of the core processing module according to an embodiment of the present invention.

[0056] Figure 9 A schematic diagram illustrating the structure of a convolutional neural network;

[0057] Figure 10 A schematic diagram illustrating the structure of a one-dimensional convolutional neural network;

[0058] Figure 11 A schematic diagram illustrating the structure of a fully connected layer;

[0059] Figure 12 This diagram illustrates the training process of the feature extraction network and the performance parameter network according to an embodiment of the present invention.

[0060] Figure 13The diagram illustrates the application process of the feature extraction network and the performance parameter network according to an embodiment of the present invention. Detailed Implementation

[0061] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.

[0062] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims.

[0063] like Figure 1 As shown, this invention provides a method and system for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion. The system comprises five parts: a data acquisition unit, a data fusion processing unit, a function implementation unit, a data storage unit, and an information display unit. The data acquisition unit is used to classify sensor signals according to their types and acquire sensor signals accordingly. The data fusion processing unit is used to preprocess the acquired sensor signals and perform multi-source heterogeneous data fusion. The function implementation unit is used to perform fault diagnosis and performance evaluation based on the results of multi-source heterogeneous data fusion, according to functional requirements. The data storage unit is used to store the results of fault diagnosis and performance evaluation. The information display unit is used to display the results of fault diagnosis and performance evaluation.

[0064] The system acquires information such as voltage, current, temperature, and audio / video data from the equipment using various types of sensors. This information is transmitted to an information display unit for visualization and to a data storage unit for storage. Then, it enters a data fusion processing unit for data fusion and processing, ultimately enabling functions such as abnormal alarms, risk warnings, and performance evaluations for equipment faults. The system's fault diagnosis and performance evaluation results are also displayed on the display unit and stored in the data storage unit. The feature extraction network and performance parameter network are modular, each corresponding to a extracted feature and a required function, facilitating expansion for different devices or needs and demonstrating strong applicability.

[0065] In one embodiment of the present invention, the system provided by the present invention is applied to single-vehicle and multi-vehicle cooperative systems, such as... Figure 2As shown, except for the vehicle speed testing unit which uses a separate Beidou remote speed measurement module, the sensors used in the other parameter testing units can be categorized into several output signal types, including DC resistance, DC current, AC voltage, DC voltage, and level signals. The sensor signal acquisition device is not only comprehensive in function but also flexible in its configuration. The system can calculate power parameters in real time based on the voltage and current parameter measurement results, realizing the measurement, conditioning, and acquisition of power parameters. For example... Figure 3 As shown, the system's hardware architecture consists of a signal adapter circuit and a general-purpose core circuit. The signal adapter circuit is specifically designed for different parameter types, completing the electrical matching of the sensor signals, converting the sensor's input signals into analog or digital signals, and sending them to the general-purpose core circuit for acquisition. For example... Figure 4 As shown, the signal adapter circuit is an ADC acquisition module. The general core circuit includes a core processing module, an AUTBUS bus processing module, a signal acquisition module, a clock module, and a power management module, which completes functions such as data acquisition, parameter storage, time synchronization and timekeeping, bus communication, and gigabit Ethernet bus communication.

[0066] The power management module supplies power to the data acquisition board and the sensors. The supply voltage includes ±13V for the data acquisition board, and -15V, +15V, +12V, +24V, and a constant current source for the sensors. The power management circuit input uses the vehicle-mounted DC power supply, with a voltage range of 10V~36V. The power supply for the data acquisition board and the sensors uses an isolated power supply design to ensure isolation and reliability between the devices. The power supply also features EMI filtering, input overvoltage and overcurrent protection, surge protection, and spike suppression. The power management circuit diagram is shown below. Figure 5 As shown.

[0067] The ADC acquisition module is connected to the core processing module and is used to convert the acquired analog signals into digital signals and output them to the core processing module. The main function of the ADC acquisition module is to convert the analog signals processed by the front-end signal adapter circuit into digital signals, and then send the converted digital signals to the ZYNQ processing unit of the core processing module via parallel data lines. A 16-bit parallel data interface is used for data transmission between the ADC and the ZYNQ. Its design block diagram is shown below. Figure 6 As shown.

[0068] The AUTBUS bus processing module, connected to the core processing module, is used for communication with the data fusion processing unit. The AUTBUS bus processing module includes: a front-end transceiver, an AUTBUS controller, and a communication interface for the core processing module, connected in sequence. The communication design with the core processing unit uses the RGMII interface, with a maximum communication rate of 1000Mbps. The RGMII interface enables data bridging from the AUTBUS bus to the core processing unit. Its design block diagram is shown below. Figure 7 As shown.

[0069] like Figure 8 As shown, the core processing module mainly includes a ZYNQ processor, eMMC storage, Ethernet interface, CAN interface, timing interface, DDR3 interface, and QSPI interface. The Ethernet and CAN interfaces enable communication with the host computer. The core processing module communicates with the fusion processing unit via the AUTBUGS bus processing module. The ZYNQ processor's PL terminal connects to the ADC acquisition module to receive the digital signals output by the ADC; the ZYNQ processor and the AUTBUGS bus processing module are connected via SGMII. The eMMC storage is connected to the ZYNQ processor's PS terminal to store the received digital signals. The timing interface connects to the clock module. The DDR3 and QSPI interfaces provide data buffering and startup program storage for the ZYNQ processor. After receiving ADC data, the ZYNQ's PL terminal processes and packages the data, then transmits it to the PS terminal via the AXI bus. The PS terminal stores the data in the eMMC storage. The DDR3 and QSPI interfaces provide data buffering and startup program storage for ZYNQ operation.

[0070] The clock module is used to provide the reference clock required for the operation of the core processing module and the AUTBUS bus processing module, including 25MHz, 100MHz, etc.

[0071] In this invention, the sensors on the equipment are divided into five categories based on their power supply method and signal output type, and are modularized so that sensors of the same type can use a single data processing module. This design makes the data acquisition unit scalable. By modularizing different types of sensors, a single data processing module can be used for sensors of the same type, increasing the generalization capability of the data acquisition unit. Furthermore, this invention uses all data as system input, avoiding the complexity of selecting input data for specific functions, and allowing for full utilization of all data for specific fault diagnosis or performance analysis.

[0072] In this embodiment of the invention, the data fusion processing unit is responsible for fusing information from signals from different sensors and extracting features to calculate equipment performance parameters, thereby obtaining the results of fault diagnosis and performance evaluation. This is mainly achieved through the following two parts: First, the collected signals are preprocessed, mainly by unifying all signal formats, denoising, and intercepting the signals to be processed for subsequent operations; Second, based on the preprocessed signals, feature-level fusion is performed using a neural network to achieve the purpose of fault anomaly warning, fault risk indication, and performance analysis.

[0073] In this embodiment of the invention, the data fusion processing unit includes a preprocessing unit and a feature extraction module. The feature extraction module includes several feature extraction networks. The feature extraction networks are used to build feature vector calculation networks for data from different sensors, and use one-dimensional convolutional neural networks (Net_cnn1D) and two-dimensional convolutional neural networks (Net_cnn2D) to extract feature vectors from one-dimensional and two-dimensional signals, respectively. A convolutional neural network (CNN) is a special type of feedforward neural network model that is better at processing images than a direct-access neural network (DNN). A typical CNN generally consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer, such as... Figure 9 As shown.

[0074] The input to a convolutional neural network is the original image. ,use Describes the features of the i-th layer of the neural network, and , The calculation process can be expressed by the formula as follows:

[0075]

[0076] in, The weight vector of the i-th convolutional kernel, denoted by " "" indicates a convolution operation. The output of the convolutional layer is related to the bias of the i-th layer. Summing and applying the summation to the activation function g(x) yields the feature map output of the i-th layer. Pooling layers are connected after convolutional layers to reduce the dimensionality of the feature maps. ReLU is used as the activation function.

[0077] like Figure 10 As shown, a one-dimensional convolutional neural network (CNN) borrows the advantages of traditional CNNs in utilizing convolution and extracting image features, using convolutional kernels of different sizes to extract feature vectors from one-dimensional signals. The one-dimensional signal is first processed by convolutional kernels of different sizes and functions to extract different features. These features are then concatenated into an initial one-dimensional feature vector, and finally, a fully connected layer is used to extract the final feature vector.

[0078] By designing targeted feature extraction network structures for different sensor signals, the signal features of different sensors are fully extracted without interference. Furthermore, this design allows for the adjustment of the entire network by simply adding a new feature extraction network for a new sensor, giving the overall algorithm structure the ability to extract features from any new data.

[0079] The functional implementation unit includes several performance parameter networks. These networks use fully connected layers to calculate the probability of different events occurring. For fault diagnosis, the calculation result is compared with a pre-set threshold to determine whether to issue an abnormal alarm or a risk warning for the fault. For performance evaluation, the calculation result informs the equipment user of the performance level. A fully connected layer (FCL) is the most basic type of neural network, consisting of an input layer, hidden layers, and an output layer. It is typically used to process one-dimensional arrays. Its structural diagram is shown below. Figure 11 As shown.

[0080] Let X be the input to the fully connected layer, and let X be the output of the i-th hidden layer. ,and . The calculation formula can be expressed as:

[0081]

[0082] in, Let b be the weight vector of the i-th hidden layer, and b be the bias. This is the activation function.

[0083] In this invention, the output range of the fully connected layer is [0,1], so the tansig function is selected as the activation function of the fully connected layer, and the logsig function is selected as the activation function of the output layer.

[0084] By designing performance parameter calculation networks tailored to different functional requirements, each function has independent performance parameter calculation results. Furthermore, this design allows for adjustments to the entire network simply by adding new performance parameter calculation networks for new functional requirements, enabling the overall algorithm structure to be extended to calculate characteristic parameters for any new function.

[0085] This invention provides a method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion, comprising the following steps:

[0086] Step S1: Acquire the signal;

[0087] Step S2: Preprocess the acquired signal, and according to the data type of the preprocessed acquired signal, use the corresponding feature extraction network to extract features from the preprocessed acquired signal, and fuse the extracted features to generate a multi-source feature vector.

[0088] Step S3: Based on the multi-source feature vectors and different analysis requirements, use the corresponding performance parameter networks to perform fault diagnosis or performance evaluation.

[0089] Step S4: Based on the increase in the data types of the acquired signals, train the basic feature extraction network to generate a new feature extraction network, and return to step S2;

[0090] Step S5: Based on the increased analysis requirements, train the basic performance parameter network to generate a new performance parameter network, and return to step S3.

[0091] In this invention, the measurement methods, principles, signal output types, and power supply methods of the sensors measuring different physical quantities are all different. They can be broadly categorized into several output signal types, such as resistance, current, AC voltage, DC voltage, and level signals. This invention aims to consider the principle and feasibility of the entire system, while also taking into account principles such as unified interfaces, independent units, and convenient expansion. Therefore, it is implemented according to the relative maximum envelope of the types of signals acquired, covering DC / AC voltage signals, DC current signals, DC resistance signals, and level signals. In subsequent use, it can cover the vast majority of sensor types, greatly improving the system's scalability while retaining the flexibility of sensor configuration.

[0092] When designing the test unit, a relatively clear understanding of the input signal type and range, as well as the power supply method, is required before subsequent design work can proceed. Based on the above design approach, the design is initially based on the following sensor signal classifications, but is not limited to these signal types. The acquisition of various signals in the solution is not limited to these; adjustments can be made according to the required sensors during subsequent implementation.

[0093] Therefore, the preprocessed acquired signals are divided into one-dimensional signal data and two-dimensional signal data. The types of one-dimensional signal data include resistance data, current data, AC voltage data, DC voltage data, level signal data, and audio signal data; the types of two-dimensional signal data include image data.

[0094] The feature extraction network for one-dimensional signal data is a one-dimensional convolutional neural network, and the feature extraction network for two-dimensional signal data is a two-dimensional convolutional neural network.

[0095] The performance parameter network is a fully connected layer, consisting of an input layer, several hidden layers, and an output layer.

[0096] Preferably, in step S2, the acquired signal is preprocessed, including:

[0097] Step S21: Perform a data integrity check on the acquired signals. The data integrity check includes handling missing items and detecting and handling outliers.

[0098] Missing value handling involves checking whether there are missing values ​​in the dataset. Missing values ​​can be handled by methods such as imputation (e.g., using the mean, median, mode, or interpolation), deletion (if the proportion of missing data is small and has little impact on the analysis results), or retention (some algorithms can handle missing values).

[0099] Outlier detection and handling: Statistical methods (such as Z-score, IQR, etc.) or model-based methods (such as cluster analysis) are used to detect outliers in the data. Outliers can be deleted, replaced with normal values, or marked so that their impact can be considered in subsequent analysis.

[0100] Step S22: Standardize the data format of the acquired signals. Data format standardization includes standardizing the time format, unit format, and data type conversion.

[0101] Among these measures, time format unification involves converting all timestamp data into a unified time format (such as UTC) to facilitate subsequent time series analysis and spatiotemporal benchmark alignment.

[0102] Unit consistency: For data involving physical quantities (such as speed, distance, temperature, etc.), ensure that all data have consistent units to avoid calculation errors caused by inconsistent units.

[0103] Data type conversion: Based on the analysis requirements, text data is converted into numerical data (e.g., using one-hot encoding, label encoding, etc.), or numerical data is converted into categorical data. The video signal is divided into a one-dimensional audio signal and a two-dimensional image sequence.

[0104] Step S23: Denoise the acquired signals, including sensor data denoising and video data denoising;

[0105] Signal denoising involves processing sensor data (such as acceleration and velocity) to eliminate the impact of sensor noise and interference signals on data analysis. Common denoising methods include filtering (such as low-pass filtering, high-pass filtering, and band-pass filtering).

[0106] Video data denoising: Denoising video data to improve the accuracy and efficiency of video analysis. Video denoising can be achieved through image denoising techniques (such as Gaussian filtering, median filtering, etc.) or by utilizing denoising algorithms from the video encoding and decoding processes.

[0107] Step S24: Determine the target interception time range based on the maximum sampling period of the acquired signal, and intercept the data of the acquired signal according to the target interception time range;

[0108] The time period of the signal to be processed should be greater than the period length of all input signals that can be analyzed, and should be an integer multiple of the maximum sampling period of all input video signals. The truncated one-dimensional signal data is the continuous signal within the target truncated time range, and the truncated two-dimensional signal data is the last image within the target truncated time range.

[0109] Step S25: Unify the data dimensions of the acquired signals.

[0110] The purpose of unifying data dimensions is to allow data of the same dimension to use the same computational model, thereby reducing the complexity of the algorithm.

[0111] Therefore, for a one-dimensional signal: the number of sampling points of the signal with the highest sampling frequency within the target intercept time range is used as the size standard, and the signal with the lower sampling rate is interpolated to the same size.

[0112] For two-dimensional signals: take the image with the highest spatial resolution as the standard, and fill the background of the low spatial resolution signal. This process should make the gray values ​​in the origin of the original image coordinate system and the eight neighboring areas the same.

[0113] Preferably, in step S25, unifying the data dimension of the one-dimensional signal includes: using the number of sampling points of the signal with the highest sampling frequency within the target intercept time range as the size standard, interpolating the acquired signals with a lower sampling frequency to the same size;

[0114] Unifying the data dimensions of two-dimensional signals includes: using the image with the highest spatial resolution as the size standard, and filling the background of signals with lower spatial resolutions to make the gray values ​​of the original image's coordinate system origin and eight neighboring regions the same.

[0115] In step S2, feature fusion is performed by vector concatenation.

[0116] In this invention, neural network technology is used to perform feature-level fusion of multi-source heterogeneous data through a feature extraction network, and to achieve fault diagnosis and performance evaluation of equipment through a performance parameter network. The neural network of this invention can be divided into two parts: a training process and an application process. The purpose of the training process is to train the model using a large number of samples to determine the parameters of each node in the neural network to form a computational model. The application process uses the trained model to achieve the purpose of fault diagnosis and performance evaluation.

[0117] Among them, such as Figure 12 As shown, the training process includes:

[0118] (1) Perform data preprocessing on the training data.

[0119] (2) Determine whether the network being trained is the initial training model or a network model with new requirements: If it is the initial training model, start training the initial feature extraction module, extract feature vectors from signals from different sensors, then fuse all feature vectors, then train the initial performance parameter module, and calculate the performance parameters for different events. After training, save all network parameters. If the initial model training has been completed, and it is a new sensor requirement or a new functional requirement, then add a corresponding feature extraction network for the data type based on the original model, or add a new performance parameter calculation network according to the new functional requirements; and train the newly added feature extraction network while keeping the parameters of the original feature extraction module unchanged, extract feature vectors from the data of the new sensor, and the data of the original sensor will be calculated by the original feature extraction module; fuse the feature vectors of the old and new sensors, and then train all performance parameter modules; after the model training is completed, save all network parameters and form a new feature extraction module and performance parameter module. If the initial model training has been completed and the new feature requirement is met, then based on the original model, a new performance parameter calculation network is added according to the feature requirement. Under the condition that the original network parameters remain unchanged, the signals from the original sensors are used to train the newly added performance parameter module. After training is completed, the network parameters are saved and a new performance parameter module is formed.

[0120] like Figure 13 As shown, the application process includes:

[0121] 1) Data preprocessing is performed on the training data. This part divides the signal into one-dimensional signal and two-dimensional signal.

[0122] 2) For data from different sensors, one-dimensional feature vectors are extracted using one-dimensional convolutional neural networks and two-dimensional convolutional neural networks respectively.

[0123] 3) Feature fusion is performed by concatenating vectors.

[0124] 4) Use fully connected layers to calculate the probability of occurrence at different times.

[0125] 5) Regarding fault issues, the abnormal alarm module outputs 1 when the probability of occurrence is greater than or equal to 0.95, and outputs 1 as a risk warning when the probability of occurrence is greater than or equal to 0.8. For performance evaluation, performance parameters are output. Abnormal alarms, risk warnings, and performance parameters are all displayed on the equipment's computer interface. The abnormal alarm and risk warning signals are converted from digital to analog signals to drive the equipment's fault light to flash and the abnormal alarm horn to issue a warning.

[0126] 6) Save the training data and the corresponding results.

[0127] In another embodiment of the present invention, feature evaluation based on neural networks can be achieved through data-level fusion. First, principal component analysis is performed on the one-dimensional data to reduce the types and dimensionality of the data to be analyzed. Second, the two-dimensional image is dimensionality reduced and fused with the one-dimensional data through methods such as singular value decomposition. Then, a deep neural network model is constructed based on the results of data-level fusion to achieve deep learning feature evaluation at the level of multi-source heterogeneous data.

[0128] In another embodiment of the present invention, 2DAE can be used as a module to form a larger deep neural network by stacking, attention mechanism and other methods to obtain the feature matrix of the two-dimensional signal. The feature vector of the one-dimensional signal can be extended to a two-dimensional matrix and fused with the feature matrix of the two-dimensional image. Finally, the fusion matrix is ​​used to construct a deep convolutional neural network to realize the evaluation of the deep convolutional neural network characteristics of multi-source heterogeneous data.

[0129] It should be noted that the above description represents a preferred embodiment of the present invention. While preferred embodiments have been described, those skilled in the art, upon understanding the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to include both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion, characterized in that, Includes the following steps: Step S1: Acquire the signal; Step S2: Preprocess the acquired signal, and according to the data type of the preprocessed acquired signal, use the corresponding feature extraction network to extract features from the preprocessed acquired signal, and fuse the extracted features to generate a multi-source feature vector. Step S3: Based on the multi-source feature vectors, and according to different analysis requirements, use the corresponding performance parameter networks to perform fault diagnosis or performance evaluation. Step S4: Based on the increase in the data types of the acquired signals, train the basic feature extraction network to generate a new feature extraction network, and return to step S2; Step S5: Based on the increased analysis requirements, train the basic performance parameter network to generate a new performance parameter network, and return to step S3.

2. The method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion according to claim 1, characterized in that, After preprocessing, the acquired signals are divided into one-dimensional signal data and two-dimensional signal data. The types of one-dimensional signal data include resistance data, current data, AC voltage data, DC voltage data, level signal data, and audio signal data; the types of two-dimensional signal data include image data. The feature extraction network for one-dimensional signal data is a one-dimensional convolutional neural network, and the feature extraction network for two-dimensional signal data is a two-dimensional convolutional neural network. The performance parameter network is a fully connected layer, consisting of an input layer, several hidden layers, and an output layer.

3. The method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion according to claim 2, characterized in that, In step S2, the acquired signal is preprocessed, including: Step S21: Perform a data integrity check on the acquired signals, including handling missing items and detecting and handling outliers; Step S22: Standardize the data format of the acquired signals. The standardization of data format includes standardization of time format, standardization of unit format, and conversion of data type. Step S23: Denoise the acquired signals, including sensor data denoising and video data denoising; Step S24: Determine the target interception time range based on the maximum sampling period of the acquired signal, and intercept the acquired signal data according to the target interception time range; the intercepted one-dimensional signal data is the continuous signal within the target interception time range, and the intercepted two-dimensional signal data is the last image within the target interception time range. Step S25: Unify the data dimensions of the acquired signals.

4. The method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion according to claim 3, characterized in that, In step S22, the data type conversion includes: According to the analysis requirements, the collected text data is converted into numerical data; Based on the analysis requirements, the collected numerical data will be converted into categorical data. The acquired video signal data is decomposed into one-dimensional audio signals and two-dimensional image sequences.

5. The method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion according to claim 3, characterized in that, In step S25, unifying the data dimension of a one-dimensional signal includes: using the number of sampling points of the signal with the highest sampling frequency within the target intercept time range as the size standard, and interpolating the acquired signals with a lower sampling frequency to the same size. Unifying the data dimensions of two-dimensional signals includes: using the image with the highest spatial resolution as the size standard, and filling the background of signals with lower spatial resolutions to make the gray values ​​of the original image's coordinate system origin and eight neighboring regions the same.

6. The method for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion according to claim 3, characterized in that, In step S2, feature fusion is performed by vector concatenation.

7. A system for acquiring and processing equipment characteristic parameter data based on multi-source heterogeneous data fusion, used to implement the equipment characteristic parameter data acquisition and processing method based on multi-source heterogeneous data fusion as described in any one of claims 1 to 6, characterized in that, include: The data acquisition unit is used to classify sensor signals according to the types of sensor signals acquired, and to acquire sensor signals according to the categories of sensor signals. A data fusion processing unit is used to preprocess the acquired sensor signals and fuse multi-source heterogeneous data; the data fusion processing unit includes a preprocessing unit and a feature extraction module, the feature extraction module including several feature extraction networks; The functional implementation unit is used to perform fault diagnosis and performance evaluation based on the fusion results of multi-source heterogeneous data according to functional requirements; the functional implementation unit includes several performance parameter networks; Data storage unit, used to store the results of fault diagnosis and performance evaluation; The information display unit is used to display the results of fault diagnosis and performance evaluation.

8. The equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion according to claim 7, characterized in that, The data acquisition unit includes a core processing module and a power management module, and also includes: An ADC acquisition module, connected to the core processing module, is used to convert the acquired analog signals into digital signals and output them to the core processing module. An AUTBUS bus processing module, connected to the core processing module, is used to communicate with the data fusion processing unit; The clock module is used to provide the reference clock required for the operation of the core processing module and the AUTBUS bus processing module.

9. The equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion according to claim 8, characterized in that, The core processing module includes: The ZYNQ processor is connected to the ADC acquisition module and is used to receive the digital signal output by the ADC acquisition module; the ZYNQ processor is also connected to the AUTBUS bus processing module. eMMC storage is used to store received digital signals; Ethernet and CAN interfaces are used for communication with the host computer. A timing interface is provided for connection to the clock module. The DDR3 and QSPI interfaces are used to provide data cache and boot program storage for the ZYNQ processor.

10. The equipment characteristic parameter data acquisition and processing system based on multi-source heterogeneous data fusion according to claim 8, characterized in that, The AUTBUS bus processing module includes: a front-end transceiver, an AUTBUS controller, and a core processing module communication interface connected in sequence; the core processing module communication interface adopts the RGMII interface.

Images