An escalator fault recognition and early warning method based on deep autoencoder
Through the escalator fault identification method based on the depth autoencoder, the fault problem caused by untimely maintenance of escalator is solved, and rapid identification and early warning is achieved, the difficulty of model establishment and understanding cost is reduced, and the accuracy and efficiency of fault identification are improved. .
Patent Information
- Application Number
- CN202210895236.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2022-05-19
- Filing Date
- 2022-07-26
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The untimely maintenance of escalators in areas with high traffic flows leads to a high probability of failure. The traditional detection methods have problems such as difficulty in establishing models and high understanding costs.
The fault identification method based on the depth autoencoder is adopted, data is collected through vibration sensors, deep autoencoder model is used for training and testing, and data visualization is combined with the t-SNE algorithm to realize escalator fault identification and early warning.
Quickly mastering the operating status of the escalator reduces the difficulty and understanding cost of model establishment, and improves the accuracy and efficiency of fault identification.
Smart Images

Figure CN115417284B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of escalator fault identification, and in particular to an escalator fault identification and early warning method based on a deep autoencoder. Background Art
[0002] With the improvement of industrialization, escalators have been widely used in schools, shopping malls, rail transit and other places. However, escalators are often installed in areas with large passenger flow, and the average daily operating time is long and the passenger volume is large. If the maintenance is not timely, the probability of escalator failure will increase.
[0003] On the basis of traditional detection methods, the application of artificial intelligence theories and methods to fault diagnosis can extract more abstract and deeper features from complex equipment operation data, and explore the complex relationship between data and equipment component failures. In particular, various tools such as deep learning and computer vision in machine learning have made breakthrough results in mechanical equipment fault identification, and have attracted the attention of more and more experts and researchers in the field. In escalator fault diagnosis, deep learning can autonomously learn valuable features in raw data and get rid of its previous method of using expert experience and manual feature extraction. Therefore, the application of deep learning-based fault detection technology to escalator fault detection research is of milestone significance. Developing intelligent diagnosis is a new way to escalator fault diagnosis. Summary of the invention
[0004] The purpose of the present invention is to provide an escalator fault identification and early warning method based on a deep autoencoder, which can identify escalator faults and facilitate managers to make decisions.
[0005] The technical solution to achieve the purpose of the present invention is:
[0006] An escalator fault recognition and early warning method based on deep autoencoder comprises the following steps:
[0007] Step 1: Data acquisition and preprocessing: The acceleration data of the escalator steps and drive device are collected through vibration sensors, the raw data are aggregated through a sliding window, and a small part of the data is labeled; the data is divided into a training set and a test set;
[0008] Step 2: Model training based on deep autoencoder: The fault recognition model based on deep autoencoder consists of three parts: encoder, decoder and classifier. First, the encoder and decoder are used to build the model, and the unlabeled training set is used to train the model weight. Based on the weight, the encoder and decoder are then used to build the model, and the model weight is trained again using a small part of the labeled training set. After the training is completed, the model is saved.
[0009] Step 3: Model test based on deep autoencoder: Input the escalator operation data to be analyzed into the model, the encoder will output the data features, and the model composed of the encoder and the classifier will output the fault identification results;
[0010] Step 4: Data visualization based on t-SNE algorithm: The data features extracted by the model are used as input to the t-SNE algorithm, which outputs two-dimensional or three-dimensional visualization results of the data features.
[0011] Compared with the prior art, the present invention has the following significant advantages:
[0012] (1) The escalator fault recognition and early warning method based on deep autoencoder of the present invention avoids the problems of difficult model establishment and high understanding cost of traditional recognition methods.
[0013] (2) The escalator fault identification and early warning method based on deep autoencoder of the present invention can enable management personnel to quickly understand the operating status of the escalator. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is a flow chart of an escalator fault identification and early warning method based on deep autoencoder.
[0015] Figure 2 Schematic diagram of the overall structure of the fault recognition model based on deep autoencoder. DETAILED DESCRIPTION
[0016] The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings.
[0017] Attached Figure 1 The present invention is a flow chart of an escalator fault identification and early warning method based on a deep autoencoder, and the present invention comprises the following steps:
[0018] Step 1, data acquisition and preprocessing: The acceleration data of the escalator steps and drive device are collected through the CTC-AC102 sensor. First, when the steps are fault-free, a vibration sensor is installed on the drive host to obtain vibration data when the drive host is fault-free, the drive host brake is poor, and the drive motor is faulty; secondly, when the drive host is fault-free, the vibration data when the steps are fault-free, the step bracket cover is broken, and the step roller is damaged are obtained; and the original data is aggregated through a sliding window with a window size of 48 and a sliding step size of 32; and a small part of the data is labeled, and the labels include the drive device is fault-free, the drive device brake is poor, the drive device motor is faulty, the steps are fault-free, the step bracket cover is broken, and the step roller is damaged, which are represented by 0-5 respectively; finally, the data is divided into a training set and a test set;
[0019] Step 2: Fault recognition model training based on deep autoencoder: Figure 2 The figure is a schematic diagram of the overall structure of the fault recognition model based on the deep autoencoder. The fault recognition model based on the deep autoencoder consists of three parts: encoder, decoder and classifier. First, the encoder and decoder are used to build the model, and the unlabeled training set is used to train the model weight. Based on the weight, the encoder and decoder are then used to build the model, and a small part of the labeled training set is used to train the model weight for the second time. After the training is completed, the model is saved.
[0020] The fault recognition model parameters based on deep autoencoder are shown in the following table:
[0021]
[0022] The network structure of the fault recognition model based on deep autoencoder is shown in the following table:
[0023]
[0024] Step 3, fault recognition model test based on deep autoencoder: input the escalator operation data to be analyzed into the model, the encoder will output the data features, and the model composed of the encoder and the classifier will output the fault recognition result; the fault recognition result is visualized through the confusion matrix, which is an effective visualization tool for realizing the performance of the classification algorithm; and the model recognition effect is evaluated according to the evaluation indicators. The commonly used indicators for measuring the model are accuracy, precision, recall and F-score; the calculation method is as follows:
[0025]
[0026]
[0027]
[0028] TP i Indicates that the data belongs to the i-th category, and the model classifies it into the i-th category; FP i Indicates that the data does not belong to the i-th category, and the model classifies it into the i-th category; FN i Indicates that the data belongs to the i-th category, and the model classifies it into other categories.
[0029] Step 4: Data visualization based on t-SNE algorithm: Input the data features extracted by the model into the t-SNE algorithm, which outputs a two-dimensional or three-dimensional visualization of the data features. The initial dimension of the data is 48, and then a deep autoencoder is used to perform preliminary dimensionality reduction on the data. After the preliminary dimensionality reduction, the data dimension is 8. Then use the t-SNE algorithm to reduce the data to 2 dimensions for visualization.
Claims
1. An escalator fault identification and early warning method based on deep autoencoder, characterized in that: The following steps are involved: Step 1: Data acquisition and preprocessing: The acceleration data of the escalator steps and drive device are collected through vibration sensors, the raw data are aggregated through a sliding window, and a small part of the data is labeled; the data is divided into a training set and a test set; Step 2: Model training based on deep autoencoder: The fault recognition model based on deep autoencoder consists of three parts: encoder, decoder and classifier. First, the encoder and decoder are used to build the model, and the unlabeled training set is used to train the model weights. Based on this weight, the encoder and decoder are used to build the model, and a small part of the labeled training set is used to train the model weights again; After training is completed, save the model; Step 3: Model test based on deep autoencoder: Input the escalator operation data to be analyzed into the model, the encoder will output the data features, and the model composed of the encoder and the classifier will output the fault identification results; Step 4: Data visualization based on t-SNE algorithm: The data features extracted by the model are used as the input of the t-SNE algorithm, and the t-SNE algorithm outputs two-dimensional or three-dimensional visualization results of the data features; Step 3: Input the escalator operation data to be analyzed into the model, the encoder will output the data features, and the model composed of the encoder and the classifier will output the fault recognition result; the fault recognition result is visualized through the confusion matrix, which is an effective visualization tool for realizing the performance of the classification algorithm; and the model recognition effect is evaluated according to the evaluation indicators, and the indicators for measuring the model are accuracy, precision, recall and F-score; the calculation method is as follows: in Indicates that the data belongs to the i-th category, and the model classifies it into the i-th category; Indicates that the data does not belong to the i-th category, and the model classifies it into the i-th category; Indicates that the data belongs to the i-th category, and the model classifies it into other categories; Step 4: Input the data features extracted by the model into the t-SNE algorithm, which outputs a two-dimensional or three-dimensional visualization of the data features. The initial dimension of the data is 48, and then the deep autoencoder is used to perform preliminary dimensionality reduction on the data. After the preliminary dimensionality reduction, the data dimension is 8. The t-SNE algorithm is then used to reduce the data to 2 dimensions for visualization.
2. The escalator fault identification and early warning method based on deep autoencoder according to claim 1 is characterized in that: Step 1: The acceleration data of the escalator steps and the drive device are collected through the CTC-AC102 sensor. First, when the steps are fault-free, a vibration sensor is installed on the drive device to obtain the vibration data when the drive device is fault-free, the drive device brake is poor, and the drive device motor is faulty. Secondly, when the drive device is fault-free, the vibration data when the steps are fault-free, the step bracket cover is broken, and the step roller is damaged are obtained. The original data is aggregated through a sliding window with a sliding window size of 48 and a sliding step size of 32. A small part of the data is labeled, and the labels include the drive device is fault-free, the drive device brake is poorly working, the drive device motor is faulty, the steps are fault-free, the step bracket cover is broken, and the step roller is damaged, which are represented by 0-5 respectively. Finally, the data is divided into a training set and a test set.
