Elevator fault judgment method and system based on big data feature analysis

Abstract

The present invention discloses a method and system for elevator fault diagnosis based on big data feature analysis. The method comprises the following steps: first, obtaining a vibration signal of an elevator car, performing preprocessing and fast Fourier transform on the vibration signal, converting the vibration signal into a feature map, gray-scaling and normalizing the feature map, and then using the processed feature map as training samples, including normal samples and fault samples, and inputting them into a CNN for learning and training. By optimizing parameters, the model has a better fault diagnosis effect, and the trained CNN is used to realize real-time judgment of the signal, so as to realize fast and accurate diagnosis of elevator faults.

Description

Technical Field

[0001] The present application belongs to the field of elevator data analysis technology, and specifically relates to an elevator fault judgment method and system based on big data feature analysis, especially an elevator fault judgment method and system based on FFT (Fast Fourier Transform) and CNN (Convolutional Neural Networks). Background Art

[0002] With the increasing popularity of elevators in recent years, people are paying more and more attention to their comfort and safety. At the same time, precisely because of this widespread use of elevators, various elevator failures are also emerging. As an important component of the entire elevator, the safety of the elevator car has a direct impact on the safety of the elevator.

[0003] The elevator car is the core component of the elevator, the part that comes into direct contact with users and serves as the central connection point for other components. Failure of other components directly affects the state of the car, especially car vibration, which directly reflects the overall operating status of the elevator. Therefore, finding a suitable method to diagnose elevator car vibration signals is of practical significance.

[0004] When an elevator malfunctions, the car generates non-stationary vibration signals during operation. These signals contain characteristic information about the elevator fault. Therefore, it is necessary to consider how to extract the required features from the vibration signals. While the commonly used method of wavelet decomposition combined with support vector machines (SVM) has shown good results, improper selection of the wavelet basis and number of decomposition layers can lead to significant errors in the results, necessitating reliance on expert experience. This makes elevator fault diagnosis based on car vibration signals currently quite difficult. Summary of the Invention

[0005] The purpose of this application is to provide an elevator fault diagnosis method and system based on big data feature analysis, which can quickly and accurately diagnose elevator faults.

[0006] To achieve the above objectives, the technical solutions adopted in this application are:

[0007] An elevator fault judgment method based on big data feature analysis, the elevator fault judgment method based on big data feature analysis comprising:

[0008] 1) Collecting the vibration signal of the elevator car and preprocessing the collected vibration signal;

[0009] 2) Convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map, including:

[0010] 21) The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m}, 1≤q≤Q;

[0011] 22) Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m};

[0012] 23) Let L(i) be the modulus of the i-th vibration signal in the data sequence Y(q), i = 1, ... m 2 ;

[0013] 24) The formula is as follows:

[0014]

[0015] Where P(j,k) is the pixel intensity of the pixel (j,k) in the converted feature map, j = 1,…m, k = 1,…m, round is the round function, Min(L) is the minimum value of the vibration signal in the data segment, and Max(L) is the maximum value of the vibration signal in the data segment;

[0016] 3) Grayscale the converted feature map to generate a grayscale image;

[0017] 4) Normalizing the values of the pixels on the grayscale image to obtain a standard image;

[0018] 5) Using the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

[0019] Several optional methods are also provided below, but they are not intended to be additional limitations on the above-mentioned overall solution. They are merely further supplements or optimizations. Under the premise that there are no technical or logical contradictions, each optional method can be combined separately for the above-mentioned overall solution, or multiple optional methods can be combined.

[0020] Preferably, the pretreatment comprises:

[0021] For missing values in the vibration signal, the missing values are filled using the mean of the vibration signals before and after the missing value position;

[0022] Use the 3σ detection method in statistics to determine outliers by calculating the standard deviation and mean of the vibration signal and eliminate them;

[0023] The vibration signal is filtered using a Butterworth filter, and the filtering formula of the Butterworth filter is:

[0024]

[0025] Where n is the order of the Butterworth filter, ω c is the cutoff frequency.

[0026] Preferably, the order of the Butterworth filter is n=1.

[0027] Preferably, the CNN is a ResNet-18 network.

[0028] The present application also provides an elevator fault diagnosis system based on big data feature analysis, which includes a signal acquisition and preprocessing module, a feature map conversion module, a grayscale conversion module, a normalization module, and a fault output module, wherein:

[0029] A signal acquisition and preprocessing module is used to acquire the vibration signal of the elevator car and preprocess the acquired vibration signal;

[0030] The feature map conversion module is used to convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map. Specifically, it performs the following operations:

[0031] The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m}, 1≤q≤Q;

[0032] Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m};

[0033] Let L(i) be the modulus value of the i-th vibration signal in the data sequence Y(q), i=1,…m 2 ;

[0034] The formula is as follows:

[0035]

[0036] Where P(j,k) is the pixel intensity of the pixel (j,k) in the converted feature map, j = 1,…m, k = 1,…m, round is the round function, Min(L) is the minimum value of the vibration signal in the data segment, and Max(L) is the maximum value of the vibration signal in the data segment;

[0037] Grayscale conversion module, used to convert the converted feature map into a grayscale image;

[0038] Normalization module, used to normalize the values of pixels on the grayscale image to obtain a standard image;

[0039] The fault output module is used to use the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

[0040] Preferably, the pretreatment comprises:

[0041] For missing values in the vibration signal, the missing values are filled using the mean of the vibration signals before and after the missing value position;

[0042] Use the 3σ detection method in statistics to determine outliers by calculating the standard deviation and mean of the vibration signal and eliminate them;

[0043] The vibration signal is filtered using a Butterworth filter, and the filtering formula of the Butterworth filter is:

[0044]

[0045] Where n is the order of the Butterworth filter, ω c is the cutoff frequency.

[0046] Preferably, the order of the Butterworth filter is n=1.

[0047] Preferably, the CNN is a ResNet-18 network.

[0048] The elevator fault diagnosis method and system based on big data feature analysis provided in the present application first obtain the vibration signal of the elevator car, preprocess and fast Fourier transform the vibration signal, convert the vibration signal into a feature map, and then grayscale and normalize the feature map. The processed feature map is then used as a training sample, including normal samples and fault samples, and input into a CNN for learning and training. By optimizing parameters, the model has a better fault diagnosis effect, and the trained CNN is used to realize real-time judgment of the signal, so as to achieve fast and accurate diagnosis of elevator faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of the elevator fault judgment method based on big data feature analysis of this application;

[0050] Figure 2 Flowchart of the convolutional neural network training for this application. DETAILED DESCRIPTION

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

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application pertains. The terms used herein in the specification of this application are for the purpose of describing specific embodiments only and are not intended to limit this application.

[0053] In order to overcome the problem of low accuracy in elevator fault diagnosis in the prior art, this embodiment proposes an elevator fault diagnosis method based on big data feature analysis.

[0054] like Figure 1 As shown, the elevator fault diagnosis method based on big data feature analysis of this embodiment includes:

[0055] 1) Collecting the vibration signal of the elevator car and preprocessing the collected vibration signal;

[0056] Vibration sensors are installed in elevator cars to collect data. Because the collected data may contain missing values and outliers, data preprocessing is required. For missing values, mean interpolation is used to fill in the missing values in the vibration signal using the mean of the data before and after the missing value. Alternatively, during CNN training, the mean of the same vibration data from different periods of time is used to fill in the missing values. For outliers, the 3σ detection method, a statistical method, is used to determine and remove outliers by calculating the standard deviation and mean of the data.

[0057] The Butterworth filter is used to filter the vibration signal containing noise. The filter formula is:

[0058]

[0059] Where n is the order of the filter, ω c is the cutoff frequency. While higher-order Butterworth filters offer better filtering performance near the cutoff frequency, they can also cause severe signal distortion, significantly impacting the accuracy of subsequent CNN learning. Therefore, in this embodiment, to ensure optimal overall filter performance, the filter order n is set to 1.

[0060] It should be noted that this embodiment uses a vibration sensor to collect the vibration signal of the elevator car. Under the premise that the signal can be collected, there is no restriction on the installation position of the vibration sensor.

[0061] 2) Convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map, including:

[0062] 21) The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m}, 1≤q≤Q.

[0063] It is easy to understand that when performing data segmentation, if the number of signals contained in the vibration signal collected this time does not meet the requirement of equal division, part of the data at the head or tail can be removed accordingly to complete the equal division.

[0064] 22) Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m}.

[0065] Perform Fast Fourier Transform on the filtered signal data to convert the signal from the original domain to the frequency domain for representation. N-1},0≤n<N, the expression of discrete Fourier transform is:

[0066]

[0067] Where e is the base of the natural logarithm and i is the imaginary unit. FFT can quickly calculate such transformations by decomposing the DFT matrix into the product of sparse factors, so it is possible to calculate the complexity of DFT (Discrete Fourier Transform). The O(n) required for the calculation is defined by DFT. 2 ) is reduced to O(nlogn).

[0068] 23) Let L(i) be the modulus of the i-th vibration signal in the data sequence Y(q), i = 1, ... m 2 .

[0069] 24) The formula is as follows:

[0070]

[0071] Where P(j,k) is the pixel intensity at pixel (j,k) in the converted feature map, j = 1,…m, k = 1,…m, round is the round function, Min(L) is the minimum vibration signal value in the data segment, and Max(L) is the maximum vibration signal value in the data segment. Using this formula, the collected vibration signal is converted into a feature map, facilitating subsequent CNN-based fault identification and improving fault identification speed.

[0072] 3) Grayscale the converted feature map to generate a grayscale image.

[0073] 4) Normalizing the values of the pixels on the grayscale image to obtain a standard image;

[0074] 5) Using the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

[0075] It's easy to understand that CNNs require appropriate training before use. That is, the fault diagnosis results they output are related to their training data. If, during training, the labels are set only to normal and faulty samples, the fault diagnosis output during real-time fault diagnosis will be either normal or faulty, determining whether the elevator is currently operating normally or faulty. If the faulty samples are further subdivided into multiple categories during training, detailed fault categories will also be output during real-time diagnosis, thus achieving the effect of multi-category elevator fault diagnosis based on the vibration signals of the elevator car.

[0076] This embodiment divides the collected vibration signal into multiple data segments for judgment. Each data segment corresponds to a different time period. The fault judgment result finally obtained for each data segment is the operating status of the elevator at different times, thereby realizing the monitoring of the elevator continuity or specified time period.

[0077] This example uses the ResNet-18 network, a classic convolutional neural network. Its characteristic is the presence of a large number of residual modules, which can avoid problems such as gradient decay or vanishing gradients. ResNet-18 includes a series of convolutional layers, identity mapping, ReLU activation function, global mean pooling, and fully connected layers.

[0078] This embodiment proposes a method for performing a fast Fourier transform on the elevator car vibration signal, converting it into a two-dimensional grayscale feature map, and then training the feature map using a convolutional neural network. Compared to other methods that directly use the vibration signal as input for neural network learning, converting the vibration signal into a two-dimensional grayscale image and feeding it into the convolutional neural network offers higher accuracy and faster convergence. Most importantly, this method's preprocessing eliminates the need for pre-defined coefficients, making it more universal.

[0079] like Figure 2 As shown, the data used to train the neural network in this embodiment comes from the elevator test tower of Hangzhou Huning Elevator Components Co., Ltd. The data used includes normal data, car fault data, traction sheave fault data, and brake guide shoe fault data, representing four operating conditions in total. The normal data is considered the fault-free data set, and the other three types of fault data are considered the fault data sets. The number of samples for each type of data is 200, and each sample has 1024 data points, for a total of 800 samples.

[0080] When training a neural network, the original samples are first preprocessed as described above, and the preprocessed samples are classified and labeled. Normal data (normal signals) are labeled as label 0, car fault data as fault signal 1 is labeled as label 1, traction sheave fault data as fault signal 2 is labeled as label 2, and brake guide shoe fault data as fault signal 3 is labeled as label 3. The data is then fast Fourier transformed and converted into a feature map. After grayscale and normalization of the feature map, samples are randomly extracted in a ratio of 5:3 and divided into training and test sets. The training set is input into the convolutional network for learning and training, and the training process is supervised. If the training results meet the expected goals (i.e., meet the accuracy requirements), the network is tested for accuracy using the data-enhanced test set. If the training results do not meet the expected goals, the training data is looped back into the network for training until the expected goals are achieved.

[0081] Since the actual operating conditions of elevators are complex and diverse, in real-time fault diagnosis, the vibration signal corresponding to the judgment result output by CNN with a confidence level lower than the preset value is uploaded to the expert system. The label of the vibration signal is determined based on expert experience, and the determined label is used for CNN training to improve CNN in real time, obtaining a judgment result with higher accuracy and more comprehensive fault diagnosis types.

[0082] In another embodiment, an elevator fault diagnosis system based on big data feature analysis is provided. The elevator fault diagnosis system based on big data feature analysis includes a signal acquisition and preprocessing module, a feature map conversion module, a grayscale conversion module, a normalization module, and a fault output module, wherein:

[0083] A signal acquisition and preprocessing module is used to acquire the vibration signal of the elevator car and preprocess the acquired vibration signal;

[0084] The feature map conversion module is used to convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map. Specifically, it performs the following operations:

[0085] The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m}, 1≤q≤Q;

[0086] Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m};

[0087] Let L(i) be the modulus value of the i-th vibration signal in the data sequence Y(q), i=1,…m 2 ;

[0088] The formula is as follows:

[0089]

[0090] Where P(j,k) is the pixel intensity of the pixel (j,k) in the converted feature map, j = 1,…m, k = 1,…m, round is the round function, Min(L) is the minimum value of the vibration signal in the data segment, and Max(L) is the maximum value of the vibration signal in the data segment;

[0091] Grayscale conversion module, used to convert the converted feature map into a grayscale image;

[0092] Normalization module, used to normalize the values of pixels on the grayscale image to obtain a standard image;

[0093] The fault output module is used to use the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

[0094] Wherein, the preprocessing includes:

[0095] For missing values in the vibration signal, the missing values are filled using the mean of the vibration signals before and after the missing value position;

[0096] Use the 3σ detection method in statistics to determine outliers by calculating the standard deviation and mean of the vibration signal and eliminate them;

[0097] The vibration signal is filtered using a Butterworth filter, and the filtering formula of the Butterworth filter is:

[0098]

[0099] Where n is the order of the Butterworth filter, ω c is the cutoff frequency.

[0100] The order of the Butterworth filter is n=1.

[0101] The CNN is a ResNet-18 network.

[0102] Regarding the limitations of the elevator fault judgment system based on big data feature analysis, please refer to the above limitations of the elevator fault judgment method based on big data feature analysis, which will not be repeated here.

[0103] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for determining elevator faults based on big data feature analysis, characterized in that: The elevator fault judgment method based on big data feature analysis includes: 1) Collecting the vibration signal of the elevator car and preprocessing the collected vibration signal; 2) Convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map, including: 21) The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m }, 1≤q≤Q; 22) Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, to obtain the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m }; 23) Let L(i) be the modulus of the i-th vibration signal in the data sequence Y(q), i=1,...m 2 ; 24) The formula is as follows: Where P(j, k) is the pixel intensity of pixel (j, k) in the converted feature map, j = 1, ... m, k = 1, ... m, round is the round function, Min(L) is the minimum value of the vibration signal in the data segment, and Max(L) is the maximum value of the vibration signal in the data segment; 3) Grayscale the converted feature map to generate a grayscale image; 4) Normalizing the values of the pixels on the grayscale image to obtain a standard image; 5) Using the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

2. The elevator fault judgment method based on big data feature analysis according to claim 1, characterized in that: The pretreatment includes: For missing values in the vibration signal, the missing values are filled using the mean of the vibration signals before and after the missing value position; Use the 3σ detection method in statistics to determine outliers by calculating the standard deviation and mean of the vibration signal and eliminate them; The vibration signal is filtered using a Butterworth filter, and the filtering formula of the Butterworth filter is: Where n is the order of the Butterworth filter, ω c is the cutoff frequency.

3. The elevator fault diagnosis method based on big data feature analysis according to claim 2, characterized in that: The order of the Butterworth filter is n=1.

4. The elevator fault diagnosis method based on big data feature analysis according to claim 1, characterized in that: The CNN is a ResNet-18 network.

5. An elevator fault diagnosis system based on big data feature analysis, characterized in that: The elevator fault diagnosis system based on big data feature analysis includes a signal acquisition and preprocessing module, a feature map conversion module, a grayscale conversion module, a normalization module, and a fault output module, wherein: A signal acquisition and preprocessing module is used to acquire the vibration signal of the elevator car and preprocess the acquired vibration signal; The feature map conversion module is used to convert the preprocessed vibration signal into a two-dimensional matrix to form a feature map. Specifically, it performs the following operations: The pre-processed vibration signal is divided into Q data segments, each data segment is represented by y(q) = {y1, y2, y3, ..., y m×m }, 1≤q≤Q; Perform FFT transformation on the data segment y(q) to obtain the modulus value, that is, the transformed data sequence Y(q) = {y1, y2, y3, ..., y m×m }; Let L(i) be the modulus value of the i-th vibration signal in the data sequence Y(q), i=1,...m 2 ; The formula is as follows: Where P(j, k) is the pixel intensity of pixel (j, k) in the converted feature map, j = 1, ... m, k = 1, ... m, round is the round function, Min(L) is the minimum value of the vibration signal in the data segment, and Max(L) is the maximum value of the vibration signal in the data segment; Grayscale conversion module, used to convert the converted feature map into a grayscale image; Normalization module, used to normalize the values of pixels on the grayscale image to obtain a standard image; The fault output module is used to use the standard image as the input of the pre-trained CNN to obtain the corresponding fault judgment result.

6. The elevator fault diagnosis system based on big data feature analysis according to claim 5, characterized in that: The pretreatment includes: For missing values in the vibration signal, the missing values are filled using the mean of the vibration signals before and after the missing value position; Use the 3σ detection method in statistics to determine outliers by calculating the standard deviation and mean of the vibration signal and eliminate them; The vibration signal is filtered using a Butterworth filter, and the filtering formula of the Butterworth filter is: Where n is the order of the Butterworth filter, ω c is the cutoff frequency.

7. The elevator fault diagnosis system based on big data feature analysis according to claim 6, characterized in that: The order of the Butterworth filter is n=1.

8. The elevator fault diagnosis system based on big data feature analysis according to claim 5, characterized in that: The CNN is a ResNet-18 network.

Images