A method and system for constructing engineering equipment health indicators

Abstract

The present invention relates to a method and system for constructing health indicators of engineering equipment, and belongs to the field of reliability engineering. The method includes: obtaining engineering equipment test data and calculating its time-frequency domain information; inputting the time-frequency domain information into a continuous deep belief network, and training it with the help of a contrast divergence algorithm; extracting deep-level fault features using the trained continuous deep belief network; using the deep-level fault features as the input of a SOM neural network for unsupervised learning; and constructing the health indicators of engineering equipment by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units. The method and system of the present invention use a continuous deep belief network to extract the deep-level fault features contained in the time-frequency domain, and use the SOM neural network to construct the health indicators of the equipment under the premise of ensuring the original feature topological structure, thereby improving the accuracy of the health indicator construction.

Description

Technical Field

[0001] The present invention relates to the field of reliability engineering technology, and in particular to a method and system for constructing health indicators of engineering equipment. Background Art

[0002] With the development of sensor monitoring technology and the improvement of electronic device technology, more and more test data can be collected with less impact on the equipment. Coupled with the increasing reliability of equipment, the amount of test data of equipment will surge. In particular, in today's era of the Internet of Things and Industry 4.0, the test data obtained by engineering equipment will show characteristics such as massive, nonlinear and dynamic. When the traditional method is used to determine the health index of the equipment, it is difficult to extract the deep fault characteristics of the equipment, which will affect the accuracy of the health index to a certain extent. In order to fully explore the hidden features contained in massive data, Hinton proposed an effective tool for analyzing massive data - deep learning, which has been highly praised by scholars and engineering technicians since then. Deep learning has been successfully applied to many engineering fields, such as image processing, speech recognition, machine translation, etc. At the same time, the fault feature extraction and health index construction based on this technology have received special attention and have achieved a series of results.

[0003] Due to the excellent characteristics and special structure of the restricted Boltzmann machine (RBM), it can be used to solve various classification, dimensionality reduction, regression and other problems. As a bidirectional probabilistic graphical model, the restricted Boltzmann machine is a neural network with a two-layer structure, mainly composed of a visible layer and a hidden layer, which meets the requirements of no connection within the layer and full connection between layers. The advantage of the restricted Boltzmann machine is that it can quantitatively represent the distribution of a set of inputs and outputs in the form of probability, which has certain advantages in processing binary data. However, when facing continuous data modeling, the model training effect is not satisfactory, which poses a challenge to the extraction of fault features of engineering equipment with continuous test data. In addition, the topological structure is another important factor that needs to be considered in health indicators. It is difficult to ensure the topological structure of the original features by simply using deep learning. Once the topological structure changes, more valuable information will be lost in the dimensionality reduction process. Therefore, it is urgent to introduce other dimensionality reduction methods to facilitate the determination of health indicators under the premise of ensuring the topological structure. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for constructing health indicators of engineering equipment, which adopts a continuous deep belief network to extract the deep fault characteristics contained in the time and frequency domains, and uses a SOM neural network to construct the health indicators of the equipment while ensuring the original feature topological structure, so as to improve the accuracy of the health indicator construction.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for constructing engineering equipment health indicators, comprising:

[0007] Acquire test data of engineering equipment and calculate time-frequency domain information of the test data;

[0008] Inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network;

[0009] Using a trained continuous deep belief network to extract deep fault features of the engineering equipment;

[0010] Using the deep fault features as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network;

[0011] By calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units, the health index of the engineering equipment is constructed; the health index directly reflects the real health status of the engineering equipment.

[0012] Optionally, acquiring test data of engineering equipment and calculating time-frequency domain information of the test data specifically includes:

[0013] Obtain test data of engineering equipment;

[0014] Extracting time domain features of the test data; the time domain features include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, shape factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor;

[0015] Extracting frequency domain features of the test data; the frequency domain features include center frequency, root mean square frequency, and frequency variance;

[0016] The extracted time domain features and the frequency domain features constitute the time and frequency domain information of the engineering equipment.

[0017] Optionally, inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network specifically includes:

[0018] Constructing a continuous deep belief network formed by stacking multiple continuous restricted Boltzmann machines; the continuous restricted Boltzmann machine includes a visual layer, a hidden layer and a Gaussian noise unit;

[0019] The time-frequency domain information is input into the continuous deep belief network, and the connection weights and sigmoid function slope control parameters of the continuous deep belief network are updated by means of the contrastive divergence algorithm to complete the continuous deep belief network parameter training and generate a trained continuous deep belief network.

[0020] Optionally, the extracting the deep fault features of the engineering equipment by using a trained continuous deep belief network specifically includes:

[0021] The time-frequency domain information X of the test data is X=[x1, x2, ..., x n ] as the input of the trained continuous deep belief network, and the output Y of the trained continuous deep belief network is Y = [y1, y2, ..., y m ] as the deep fault characteristics of the engineering equipment; where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of input features; y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

[0022] Optionally, the step of using the deep-level fault features as input of a SOM neural network for unsupervised learning to generate a trained SOM neural network specifically includes:

[0023] The deep fault feature Y=[y1,y2,…,y m ] is used as the input vector D of the SOM neural network, and the weight vectors of the best matching unit and the topological neighborhood are updated according to the input vector D to complete the unsupervised learning of the SOM neural network and generate a trained SOM neural network.

[0024] A system for constructing engineering equipment health indicators, comprising:

[0025] A time-frequency domain information extraction module, used to obtain test data of engineering equipment and calculate the time-frequency domain information of the test data;

[0026] A continuous deep belief network training module, used for inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network;

[0027] A deep fault feature extraction module, used to extract the deep fault features of the engineering equipment using a trained continuous deep belief network;

[0028] A SOM neural network training module, used for performing unsupervised learning by taking the deep fault features as input of the SOM neural network to generate a trained SOM neural network;

[0029] The engineering equipment health index construction module is used to construct the health index of the engineering equipment by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units; the health index directly reflects the actual health status of the engineering equipment.

[0030] Optionally, the time-frequency domain information extraction module specifically includes:

[0031] A test data acquisition unit, used to acquire test data of engineering equipment;

[0032] A time domain feature extraction unit, used to extract the time domain features of the test data; the time domain features include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, waveform factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor;

[0033] A frequency domain feature extraction unit, used to extract the frequency domain features of the test data; the frequency domain features include center frequency, root mean square frequency, and frequency variance;

[0034] The time-frequency domain information construction unit is used to construct the time-frequency domain information of the engineering equipment by using the extracted time-domain features and the extracted frequency-domain features.

[0035] Optionally, the continuous deep belief network training module specifically includes:

[0036] A continuous deep belief network construction unit, used to construct a continuous deep belief network formed by stacking multiple continuous restricted Boltzmann machines; the continuous restricted Boltzmann machine includes a visible layer, a hidden layer and a Gaussian noise unit;

[0037] A continuous deep belief network training unit is used to input the time-frequency domain information into the continuous deep belief network, update the connection weights and sigmoid function slope control parameters of the continuous deep belief network by means of a contrastive divergence algorithm, complete the continuous deep belief network parameter training, and generate a trained continuous deep belief network.

[0038] Optionally, the deep fault feature extraction module specifically includes:

[0039] A deep fault feature extraction unit is used to extract the time-frequency domain information X of the test data from the n] as the input of the trained continuous deep belief network, and the output Y of the trained continuous deep belief network is Y = [y1, y2, ..., y m ] as the deep fault characteristics of the engineering equipment; where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of input features; y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

[0040] Optionally, the SOM neural network training module specifically includes:

[0041] The SOM neural network training unit is used to transform the deep fault feature Y=[y1,y2,…,y m ] is used as the input vector D of the SOM neural network, and the weight vectors of the best matching unit and the topological neighborhood are updated according to the input vector D to complete the unsupervised learning of the SOM neural network and generate a trained SOM neural network.

[0042] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] The present invention provides a method and system for constructing health indicators of engineering equipment, the method comprising: obtaining test data of engineering equipment and calculating time-frequency domain information of the test data; inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrast divergence algorithm, and generating a trained continuous deep belief network; extracting deep-level fault features of the engineering equipment using the trained continuous deep belief network; using the deep-level fault features as input of a SOM neural network for unsupervised learning, and generating a trained SOM neural network; constructing the health indicator of the engineering equipment by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units; the health indicator directly reflects the true health status of the engineering equipment. The method and system of the present invention use a continuous deep belief network to extract the deep-level fault features contained in the time-frequency domain, and use the SOM neural network to construct the health indicator of the equipment under the premise of ensuring the original feature topological structure, thereby improving the accuracy of the health indicator construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A flowchart of a method for constructing a health indicator of engineering equipment according to the present invention;

[0046] Figure 2 A schematic diagram of a bearing experimental platform provided in an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of a typical horizontal vibration signal of a bearing provided in an embodiment of the present invention during its entire life cycle;

[0048] Figure 4 A schematic diagram of health indicators of a bearing under the first operating condition of the three operating conditions provided in an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of health indicators of a bearing under the second operating condition of the three operating conditions provided in an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of health indicators of a bearing under the third operating condition of the three operating conditions provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Aiming at engineering equipment under the background of big data, the present invention proposes a universal method and system for fault feature extraction and health index construction. A continuous deep belief network is used to extract the deep fault features contained in the time and frequency domains, and a SOM neural network is used to construct the health index of the equipment under the premise of ensuring the original feature topological structure, so as to improve the characteristics of the health index and improve the accuracy of the health index construction.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] Figure 1This is a flow chart of a method for constructing engineering equipment health indicators according to the present invention. Figure 1 , a method for constructing engineering equipment health indicators of the present invention includes:

[0055] Step 1: Obtain test data of engineering equipment and calculate time-frequency domain information of the test data.

[0056] The present invention uses signal processing technology to analyze the collected massive test data to determine the corresponding time domain and frequency domain characteristics. Typical time domain characteristics mainly include mean, root mean square, root amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, waveform factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor; frequency domain characteristics mainly include center frequency, root mean square frequency, frequency variance, etc.

[0057] Therefore, the step 1 of acquiring the test data of the engineering equipment and calculating the time-frequency domain information of the test data specifically includes:

[0058] Step 1.1: Obtain test data of engineering equipment.

[0059] The engineering equipment includes mechanical equipment such as bearings, as well as electronic equipment such as chips, transistors, and MOS tubes.

[0060] Step 1.2: Extract the time domain characteristics of the test data; the time domain characteristics include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, waveform factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor.

[0061] The calculation formulas for each time domain feature are as follows:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] Maximum x max =maxx(t) (8)

[0070] Minimum value x min=minx(t) (9)

[0071] Peak x peak =max|x(t)| (10)

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] Among them, x i represents the i-th test data of the engineering equipment in a single sampling interval; N is the number of test data in a single sampling interval, and x(t) represents the set of all data of the engineering equipment in a single sampling interval.

[0079] Step 1.3: Extract the frequency domain features of the test data; the frequency domain features mainly include center frequency, root mean square frequency, frequency variance, etc. To facilitate the understanding of frequency domain features, the calculation formula is described in detail using center frequency, root mean square frequency, and frequency variance as examples:

[0080]

[0081]

[0082]

[0083] Performing Fourier transform on the collected engineering equipment test data can extract the contained spectrum information from the time domain signal, that is, the frequency value and the corresponding peak value / amplitude. The frequency domain features can be calculated based on the spectrum information. i is the frequency value corresponding to the spectrum at the i-th moment, p i represents the amplitude in the frequency domain corresponding to the i-th moment, f c is the average frequency. Other frequency domain features can be determined based on existing literature and combined with the above frequency domain features.

[0084] Step 1.4: The extracted time domain features and the frequency domain features constitute the time and frequency domain information of the engineering equipment.

[0085] The time-frequency domain features of each monitoring moment are combined into a feature vector. For multiple monitoring moments, different feature vectors are combined into a feature matrix in chronological order, that is, the time-frequency domain information of the engineering equipment is constituted as the input of the continuous deep belief network.

[0086] Step 2: Input the time-frequency domain information into a continuous deep belief network, train the continuous deep belief network with the help of a contrastive divergence algorithm, and generate a trained continuous deep belief network.

[0087] The present invention inputs the extracted time-frequency domain information into a continuous deep belief network (CDBN), and trains the network model with the help of a contrastive divergence algorithm. After the training is completed, the deep-level fault characteristics of the engineering equipment can be extracted.

[0088] The step 2 inputs the time-frequency domain information into a continuous deep belief network, trains the continuous deep belief network by means of a contrastive divergence algorithm, and generates a trained continuous deep belief network, specifically including:

[0089] Step 2.1: Construct a continuous deep belief network composed of multiple continuous restricted Boltzmann machines stacked together; the continuous restricted Boltzmann machine includes a visible layer, a hidden layer and a Gaussian noise unit.

[0090] The continuous deep belief network constructed by the present invention is a neural network composed of multiple continuous restricted Boltzmann machines (CRBM). The continuous restricted Boltzmann machine is based on the traditional restricted Boltzmann machine and is constructed by adding a zero-mean Gaussian noise to the sampled Sigmoid unit. It mainly includes a visual layer, a hidden layer and a Gaussian noise unit. Let s j Represents a set of neuron states {s i} is the output state of neuron j, then s j It can be expressed as:

[0091]

[0092] Among them, W ij is the connection weight, N j (0,1) represents Gaussian white noise with mean 0 and variance 1. σ is a constant term, which is related to N j (0,1) together constitute the noise input part, that is, n j =σ·N j (0,1), the corresponding probability density density p(n j )for:

[0093]

[0094] φ j (x j ) is a more general sigmoid function that satisfies x j is the state before the sigmoid function is applied, satisfying θ L With θ H Respectively represent φ j (x j ) The lower and upper asymptotes of the function, c j is a parameter that controls the slope of the sigmoid function, called the sigmoid function slope control parameter.

[0095] Step 2.2: Input the time-frequency domain information into the continuous deep belief network, update the connection weights and sigmoid function slope control parameters of the continuous deep belief network by means of the contrastive divergence algorithm, complete the continuous deep belief network parameter training, and generate a trained continuous deep belief network.

[0096] {s in step 2.1 i} represents each neuron in the input layer of the continuous restricted Boltzmann machine. In order to conveniently describe the operation mechanism of the continuous restricted Boltzmann machine, {s i} indicates that when the time-domain frequency-domain features (i.e., the time-domain frequency-domain information) are input into the continuous deep belief network, the first continuous restricted Boltzmann machine input layer directly receives the time-domain frequency-domain features. Therefore, {s i} needs to be replaced by time domain and frequency domain features, but for the subsequent continuous restricted Boltzmann machine, {s i} needs to be replaced by the output of the previous continuous restricted Boltzmann machine.

[0097] When the neurons in the visual layer input a specific state, the state of the hidden layer neurons can be obtained according to formula (20), and the state of the visual layer can be reconstructed based on the state of the hidden layer neurons. The state of the hidden layer neurons after reconstruction can be determined in the same way. By analogy, after infinite reconstructions, the training of model parameters can be achieved. However, this method also requires multiple Gibbs samplings, which is time-consuming and not conducive to real-time solution of the model. The present invention adopts a minimization contrast divergence algorithm for model parameter training, which significantly reduces the amount of calculation and provides a convenient way for efficient training of model parameters. The continuous deep belief network weights and the sigmoid function slope control parameters can be updated according to the following formula:

[0098]

[0099]

[0100] In the formula, ε w With ε c Represent the network connection weight W ij and the sigmoid function slope control parameter c j The learning rate. s i s j is the product of the current states of neuron i and neuron j. The state values ​​of neurons i and j are sampled one step at a time. For a single continuous restricted Boltzmann machine, when the total number of neurons in the hidden layer is less than the total number of neurons in the visual layer, the dimension of the input state can be effectively reduced. ΔW ij It means that after one sampling step, W ij The change in Δc j It means that after one sampling step c j The change in ΔW can be realized after multiple sampling steps. ij and Δc j Respectively with the parameters W of the previous training ij and c j The parameters are updated by adding them together, thereby completing the continuous deep belief network parameter training and generating a trained continuous deep belief network.

[0101] Step 3: Use the trained continuous deep belief network to extract the deep fault features of the engineering equipment.

[0102] Assuming that the original characteristics of the test signal of the degraded device, that is, the time-frequency domain information of the test data extracted can be expressed as X = [x1, x2, ..., x n ], where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of original features. These original features are used as the input of the continuous deep belief network. The model parameters of each continuous restricted Boltzmann machine are determined layer by layer using the above-mentioned contrast divergence algorithm. Based on formula (20), the state of each layer of neurons in the continuous deep belief network can be calculated, that is, the deep fault features reflecting the health status of the equipment can be extracted through multiple continuous restricted Boltzmann machines. Therefore, the output of the continuous deep belief network can be regarded as the deep fault features of the input data, which can be expressed as Y=[y1,y2,…,y m ], where y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

[0103] The time-frequency domain information of the test data X = [x1, x2, ..., x n] contains the time domain and frequency domain features of all data, i represents the feature number, x i (0≤i≤n) is the sequence value of the ith feature. Because it is sampled at multiple time points, it is a sequence. Each feature directly corresponds to a neuron in the continuous deep belief network input. Each row in X represents a vector composed of time domain and frequency domain features at a certain moment, and each column represents the value of a certain feature at different moments. Therefore, after X is input into the continuous deep belief network, each row of X corresponds to the input neuron of the first continuous restricted Boltzmann machine of the continuous deep belief network, that is, {s i} is corresponding, and for the subsequent continuous restricted Boltzmann machine, {s i} only represents the output of the previous layer of continuous restricted Boltzmann machine.

[0104] Step 4: Use the deep fault features as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network.

[0105] The present invention uses the fault features extracted by the continuous deep belief network as the input of the SOM (Self-organizing map) neural network to achieve its unsupervised learning, and quantitatively characterizes the health status of the equipment by calculating the Euclidean distance between the state vector and the weight vectors of all units.

[0106] The SOM neural network constructed by the present invention mainly includes two parts: input layer and output layer. The input layer mainly receives actual medium and high-dimensional nonlinear data, also known as the competition layer; the output layer is composed of a regular node grid. The weight vector between the lth neuron in the output layer and the input neuron can be expressed as W l =[W l1 ,W l2 ,…,W lm ], m is the dimension of the SOM neural network input vector. For a row of the deep fault feature Y extracted in step 3, that is, the deep feature from the continuous deep belief network at a specific moment can be regarded as the SOM neural network input vector, which can be recorded as D. According to the input vector D, the weight vector of the best matching unit and the topological neighborhood is updated to complete the unsupervised learning of the SOM neural network, thereby generating a trained SOM neural network.

[0107] The learning process mainly calculates the Euclidean distance between the input vector and the weight vector to determine the neuron closest to the input vector in each step of training. The closest neuron is generally called the Best Matching Unit (BMU). Based on the obtained BMU, the weight vector of the BMU and the topological neighborhood can be adjusted and updated. The update result can be expressed as:

[0108]

[0109] Where p is the index of the iteration step, W l (p) and W l (p+1) are the weight vectors of p step and p+1 step respectively, D is the input vector, ε(p) is the learning rate, represents the topological neighborhood kernel function centered on BMU, W BMU is the weight vector of BMU.

[0110] Step 5: The health index of the engineering equipment is constructed by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units.

[0111] In the process of health assessment, the SOM structure is trained using health status data. After the training is completed, the state vector of the input layer is compared with the weight vectors of all units to evaluate the health status of the device. Specifically, the present invention selects the minimum quantization error as the health indicator, that is, by calculating the Euclidean distance between the state vector and the weight vectors of all units, the health status of the device is quantitatively characterized, that is:

[0112] Z=min||DW l || (25)

[0113] Since the input vectors at different times correspond to different health indicators, a series of health indicator sequences Z = [Z(t1), Z(t2), …, Z(t N )], where Z(t i )(1≤i≤N) represents t i It should be noted that the SOM neural network can convert multidimensional data into a one-dimensional health indicator, which directly reflects the true health status of the equipment. Fault diagnosis and life prediction can be performed based on this health indicator.

[0114] The health index sequence Z extracted by the present invention is [Z(t1), Z(t2), ..., Z(t N )] is a quantity that changes with time. If this indicator has an upward trend over time, it means that the equipment has a degradation trend, indicating that the health status will become worse and worse. This health indicator is a comprehensive indicator extracted by fusing all the time domain and frequency domain features of the equipment with the help of a continuous deep belief network and a SOM neural network, so that the original features can be expressed in an abstract way. The specific implementation example of the bearing below can reflect this process.

[0115] The following uses the public bearing experimental data to verify the effectiveness and superiority of the method. The bearing experimental data set comes from the joint laboratory for mechanical equipment health monitoring jointly established by the Institute of Design Science and Basic Components of Xi'an Jiaotong University and Changxing Shengyang Technology Co., Ltd. The bearing experimental platform is mainly composed of an AC motor, a motor speed controller, a support shaft, two support bearings, and a hydraulic loading device, such as Figure 2 As shown. The sensor device of the test bench is two accelerometers installed vertically on the outer ring of the experimental bearing. Its main function is to collect vibration signals in the horizontal and vertical directions. The single sampling frequency is 25.6kHz, the sampling duration is 1.28s, and the interval between two adjacent samplings is 1min.

[0116] The accelerated bearing degradation experiment was carried out under three experimental conditions. Five groups of bearings were tested under each experimental condition. Therefore, the data set mainly includes the full life cycle vibration test data of 15 groups of bearings. Table 1 lists the specific information on the operating conditions and life of the experimental bearings.

[0117] Table 1 Experimental bearing operating conditions and life details

[0118]

[0119]

[0120] Since the experimental load is applied in the horizontal direction, the accelerometer installed in the horizontal direction can obtain more bearing degradation information. Therefore, the horizontal vibration signal is selected here to adaptively predict the RUL of the test bearing. Figure 3 The typical horizontal vibration signal of a bearing over its entire life cycle is depicted. Figure 3 It can be seen intuitively that the health status of the bearing can usually be divided into two stages: stable operation stage and degradation stage. In the stable operation stage, the vibration signal of the bearing fluctuates slightly around the value of 0, and there is no degradation trend, so the degradation of the bearing in this stage can be ignored; however, in the degradation stage, the amplitude of the vibration signal has a significant upward trend over time, which means that extremely rich bearing degradation information can be obtained in this stage.

[0121] In order to make full use of the characteristic information in the vibration signal, the 16 time domain features and 13 frequency domain features described in step 1.2 (including the center frequency, root mean square frequency, frequency variance and the remaining 10 other frequency domain features well known in the art described in step 1.3) are selected here and input into the CDBN containing a 2-layer CRBM structure to extract the deep fault features contained in the vibration signal. The specific structure and initial parameters of the CDBN and SOM neural networks are determined by cross-validation, as listed in Tables 2 and 3.

[0122] Table 2 Specific structure and initial parameters of CDBN

[0123]

[0124]

[0125] Table 3 Specific structure and initial parameters of SOM

[0126] structure Topological form Initial number of neighbors Settings 23*23 Hextop 3

[0127] After extracting the implicit deep fault features reflecting the health status of the bearing based on CDBN, the SOM neural network can be used to determine the health indicators of the bearing under three operating conditions, as follows: Figure 4 , Figure 5 , Figure 6 As shown in Figure 2, the 29 features will be simplified to one-dimensional features. Figure 4 It shows that the trend of the bearing health index is consistent with the amplitude trend of the vibration signal, which confirms the two-stage health state of the bearing to a certain extent. Therefore, the method of the present invention is reasonable and effective in extracting the deep fault characteristics of the bearing and determining the health index.

[0128] In order to further demonstrate the superiority of the method proposed in the present invention that combines the continuous deep belief network and the SOM neural network, traditional dimensionality reduction methods are used as comparison methods to determine the health indicators, such as PCA (Principle Component Analysis), SOM neural network and continuous deep belief network (CDBN). At the same time, the health indicator trend factor is selected to evaluate the quality of the health indicator. Generally speaking, the closer the trend factor is to 1, the stronger the trend. Table 4 lists the trend factor results of each bearing health indicator under the proposed method and the three comparison methods.

[0129] Table 4 Comparison of health index performance under four methods

[0130]

[0131]

[0132] From the data in Table 4, it can be found that the trend factor of the health index calculated by the method of the present invention is greater than the trend factors of the health index under the other three methods, which means that the health index determined by the method of the present invention is better than the three traditional dimensionality reduction methods. In addition, the trend factors of most bearing health indicators determined by the method of the present invention are greater than 0.5, which provides technical support for the degradation modeling of health indicators and RUL (Remaining Useful Life) prediction.

[0133] The present invention proposes a new method for constructing engineering equipment health indicators. For engineering equipment under the background of big data, it can extract the deep-level fault characteristics contained in the time-frequency domain, and construct the equipment health indicators under the premise of ensuring the original feature topological structure. Compared with the prior art, the method of the present invention fully extracts the deep-level fault characteristics contained in the time-frequency domain information, while ensuring the topological structure of the original features in the dimensionality reduction process, further develops the fault feature extraction method, enriches the theoretical system of equipment health indicator construction, and effectively improves the performance characteristics of health indicators.

[0134] Based on a method for constructing an engineering equipment health indicator provided by the present invention, the present invention also provides a system for constructing an engineering equipment health indicator, the system comprising:

[0135] A time-frequency domain information extraction module, used to obtain test data of engineering equipment and calculate the time-frequency domain information of the test data;

[0136] A continuous deep belief network training module, used for inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network;

[0137] A deep fault feature extraction module, used to extract the deep fault features of the engineering equipment using a trained continuous deep belief network;

[0138] A SOM neural network training module is used to use the deep fault characteristics as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network;

[0139] The engineering equipment health index construction module is used to construct the health index of the engineering equipment by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units; the health index directly reflects the actual health status of the engineering equipment.

[0140] The time-frequency domain information extraction module specifically includes:

[0141] A test data acquisition unit, used to acquire test data of engineering equipment;

[0142] A time domain feature extraction unit, used to extract the time domain features of the test data; the time domain features include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, waveform factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor;

[0143] A frequency domain feature extraction unit, used to extract the frequency domain features of the test data; the frequency domain features include center frequency, root mean square frequency, and frequency variance;

[0144] The time-frequency domain information construction unit is used to construct the time-frequency domain information of the engineering equipment by using the extracted time-domain features and the extracted frequency-domain features.

[0145] The continuous deep belief network training module specifically includes:

[0146] A continuous deep belief network construction unit, used to construct a continuous deep belief network formed by stacking multiple continuous restricted Boltzmann machines; the continuous restricted Boltzmann machine includes a visible layer, a hidden layer and a Gaussian noise unit;

[0147] A continuous deep belief network training unit is used to input the time-frequency domain information into the continuous deep belief network, update the connection weights and sigmoid function slope control parameters of the continuous deep belief network by means of a contrastive divergence algorithm, complete the continuous deep belief network parameter training, and generate a trained continuous deep belief network.

[0148] The deep fault feature extraction module specifically includes:

[0149] A deep fault feature extraction unit is used to extract the time-frequency domain information X=[x1, x2,…, x n ] as the input of the trained continuous deep belief network, and the output Y of the trained continuous deep belief network is Y = [y1, y2, ..., y m ] as the deep fault characteristics of the engineering equipment; where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of input features; y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

[0150] The SOM neural network training module specifically includes:

[0151] The SOM neural network training unit is used to transform the deep fault feature Y=[y1,y2,…,y m ] is used as the input vector D of the SOM neural network, and the weight vectors of the best matching unit and the topological neighborhood are updated according to the input vector D to complete the unsupervised learning of the SOM neural network and generate a trained SOM neural network.

[0152] The present invention provides a novel method and system for constructing engineering equipment health indicators, which fully extracts the deep fault characteristics contained in the time and frequency domains, while ensuring the topological structure of the original characteristics in the dimensionality reduction process. First, the test data of the engineering equipment is collected and the corresponding time and frequency domain information is calculated; then, the time and frequency domain information is input into the continuous deep belief network, and the network model is trained with the help of the contrast divergence algorithm. After the training is completed, the deep fault characteristics of the engineering equipment can be extracted, and the fault characteristics are used as the input of the SOM neural network to realize its unsupervised learning. By calculating the Euclidean distance between the state vector and the weight vectors of all units, the health status of the equipment is quantitatively characterized. The method described in the present invention further develops the fault feature extraction method, enriches the theoretical system for constructing equipment health indicators, effectively improves the performance characteristics of health indicators, and is helpful for subsequent fault diagnosis and life prediction, and has a broad application space.

[0153] For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0154] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for constructing engineering equipment health indicators, characterized in that: include: Acquire test data of engineering equipment and calculate time-frequency domain information of the test data; Inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network; Using a trained continuous deep belief network to extract deep fault features of the engineering equipment; Using the deep fault features as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network; By calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units, the health index of the engineering equipment is constructed; the health index directly reflects the real health status of the engineering equipment.

2. The method according to claim 1, characterized in that The obtaining of the test data of the engineering equipment and calculating the time-frequency domain information of the test data specifically includes: Obtain test data of engineering equipment; Extracting time domain features of the test data; the time domain features include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, shape factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor; Extracting frequency domain features of the test data; the frequency domain features include center frequency, root mean square frequency, and frequency variance; The extracted time domain features and the frequency domain features constitute the time and frequency domain information of the engineering equipment.

3. The method according to claim 2, characterized in that The step of inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network specifically includes: Constructing a continuous deep belief network formed by stacking multiple continuous restricted Boltzmann machines; the continuous restricted Boltzmann machine includes a visual layer, a hidden layer and a Gaussian noise unit; The time-frequency domain information is input into the continuous deep belief network, and the connection weights and sigmoid function slope control parameters of the continuous deep belief network are updated by means of the contrastive divergence algorithm to complete the continuous deep belief network parameter training and generate a trained continuous deep belief network.

4. The method according to claim 3, characterized in that The method of extracting the deep fault features of the engineering equipment by using the trained continuous deep belief network specifically includes: The time-frequency domain information X of the test data is X=[x1, x2, ..., x n ] as the input of the trained continuous deep belief network, and the output Y of the trained continuous deep belief network is Y = [y1, y2, ..., y m ] as the deep fault characteristics of the engineering equipment; where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of input features; y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

5. The method according to claim 4, characterized in that The method of using the deep fault features as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network specifically includes: The deep fault feature Y=[y1,y2,…,y m ] is used as the input vector D of the SOM neural network, and the weight vectors of the best matching unit and the topological neighborhood are updated according to the input vector D to complete the unsupervised learning of the SOM neural network and generate a trained SOM neural network.

6. A system for constructing engineering equipment health indicators, characterized in that: include: A time-frequency domain information extraction module, used to obtain test data of engineering equipment and calculate the time-frequency domain information of the test data; A continuous deep belief network training module, used for inputting the time-frequency domain information into a continuous deep belief network, training the continuous deep belief network by means of a contrastive divergence algorithm, and generating a trained continuous deep belief network; A deep fault feature extraction module, used to extract the deep fault features of the engineering equipment using a trained continuous deep belief network; A SOM neural network training module is used to use the deep fault characteristics as input of the SOM neural network for unsupervised learning to generate a trained SOM neural network; The engineering equipment health index construction module is used to construct the health index of the engineering equipment by calculating the Euclidean distance between the state vector of the input layer of the trained SOM neural network and the weight vectors of all units; the health index directly reflects the actual health status of the engineering equipment.

7. The system according to claim 6, characterized in that The time-frequency domain information extraction module specifically includes: A test data acquisition unit, used to acquire test data of engineering equipment; A time domain feature extraction unit, used to extract the time domain features of the test data; the time domain features include mean, root mean square, root square amplitude, average absolute amplitude, skewness, kurtosis, variance, maximum value, minimum value, peak value, waveform factor, peak factor, pulse factor, margin factor, skewness factor, and kurtosis factor; A frequency domain feature extraction unit, used to extract the frequency domain features of the test data; the frequency domain features include center frequency, root mean square frequency, and frequency variance; The time-frequency domain information construction unit is used to construct the time-frequency domain information of the engineering equipment by using the extracted time-domain features and the extracted frequency-domain features.

8. The system according to claim 7, characterized in that The continuous deep belief network training module specifically includes: A continuous deep belief network construction unit, used to construct a continuous deep belief network formed by stacking multiple continuous restricted Boltzmann machines; the continuous restricted Boltzmann machine includes a visible layer, a hidden layer and a Gaussian noise unit; A continuous deep belief network training unit is used to input the time-frequency domain information into the continuous deep belief network, update the connection weights and sigmoid function slope control parameters of the continuous deep belief network by means of a contrastive divergence algorithm, complete the continuous deep belief network parameter training, and generate a trained continuous deep belief network.

9. The system according to claim 8, characterized in that The deep fault feature extraction module specifically includes: A deep fault feature extraction unit is used to extract the time-frequency domain information X=[x1, x2,…, x n ] as the input of the trained continuous deep belief network, and the output Y of the trained continuous deep belief network is Y = [y1, y2, ..., y m ] as the deep fault characteristics of the engineering equipment; where x i (0≤i≤n) represents the feature sequence of the i-th input neuron of the continuous deep belief network, and n is the total number of input features; y i (0≤i≤m) represents the feature sequence of the i-th output neuron of the continuous deep belief network, and m is the total number of output neurons of the continuous deep belief network.

10. The system according to claim 9, characterized in that The SOM neural network training module specifically includes: The SOM neural network training unit is used to transform the deep fault feature Y=[y1,y2,…,y m ] is used as the input vector D of the SOM neural network, and the weight vectors of the best matching unit and the topological neighborhood are updated according to the input vector D to complete the unsupervised learning of the SOM neural network and generate a trained SOM neural network.

Images