CN113987905A - Escalator braking force intelligent diagnosis system based on deep belief network - Google Patents

Abstract

The invention discloses an escalator braking force intelligent diagnosis system based on a deep belief network, which comprises a braking distance acquisition module, a braking amplitude acquisition module, a braking performance database and a monitoring module, wherein the braking distance acquisition module acquires braking distance, braking amplitude and escalator technical parameters and uploads the data to the braking performance database; B. building an escalator braking force prediction model and building an escalator braking force fault diagnosis model through deep learning; C. the brake performance database stores and reduces the data; D. comparing the reduced data with a braking force prediction model and a braking force fault diagnosis model; E. and checking real-time data and early warning through the braking force intelligent diagnosis platform according to the comparison result. The escalator brake force early warning method has the advantages that the brake force range of the escalator is estimated by using the deep belief network, useful fault characteristics are directly extracted from escalator fault original data, fault states are identified, early warning of brake force abnormity and diagnosis results of brake force faults can be obtained in the first time, hidden dangers existing in the escalator are timely checked, intelligent management is achieved, and the operation efficiency of the escalator is improved.

Description

An intelligent diagnosis system for escalator braking force based on deep belief network

technical field

The invention relates to an escalator braking force detection system, in particular to an escalator braking force intelligent diagnosis system based on a deep belief network.

Background technique

As a special equipment for continuous transportation of people, escalators and moving sidewalks have become important transportation tools in public places due to their stable operation, safety and reliability, strong transportation capacity, and convenient use. They are widely used in airports, stations, shopping malls, supermarkets, hospitals and other public places. At the same time, escalator accidents also occur from time to time due to the large amount of use, frequent use, large passenger capacity, and complex use environment. These accidents often have the characteristics of many people involved and serious injuries. Safe operation has become a new research direction.

Through the investigation and analysis of escalator and moving sidewalk accident cases, it is found that there are mainly the following reasons for the accident. Table 1 The cause of the accident and its proportion.

Although the number of escalator reversal accidents is not the most, its harm is characterized by group nature and strong social repercussions, and it usually occurs on escalators that are fully loaded. The mechanical failures that cause the reversal of the escalator are: drive chain breakage, step chain breakage and insufficient braking torque of the working brake. Because the safety margin is fully considered in the design of the drive chain and step chain, the probability of breakage is small, and the brake wheel and the brake shoe of the escalator will produce friction when the escalator stops, resulting in the wear of the brake shoe. Improper braking will result in insufficient braking power, which will greatly increase the risk of reversal. At present, in the maintenance and regular inspection of the escalator, the braking distance of the escalator is usually measured to indirectly verify its braking ability, but the braking distance includes the measurement results under two conditions of load and no load. The no-load test is carried out, and the on-load test is carried out only after the supervision inspection, and the inspection frequency is not high, and many hidden safety hazards cannot be found in time during the inspection. In addition, the factors that affect the stopping distance of the escalator include the lubrication of moving parts such as step chains and sprockets, and the performance of braking force. It is insufficient to judge whether the braking force is qualified from the braking distance alone.

The existing escalator online detection system can monitor in real time and remotely understand the use and operation rules of escalators; it has a certain ability of escalator braking force early warning, but it cannot detect regular potential risks in time, and there is a danger of risk expansion. , without intelligent diagnosis of escalator braking force. Under the current industry trend of "Internet" + "special equipment detection", the research on online monitoring of special equipment and intelligent early warning technology is gradually improving in the elevator field. Therefore, it is very important to study an intelligent diagnosis system for escalator braking force.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a device that can identify abnormal changes in braking force, provide technical support for maintenance personnel to maintain and repair escalators, and integrate real-time monitoring, data statistics, intelligent early warning, and fault diagnosis into one, and improve the Escalator braking force intelligent diagnosis system for escalator operating efficiency.

The technical solution adopted in the present invention is an intelligent diagnosis system for escalator braking force based on a deep belief network, comprising the steps of:

A. Collection of braking distance, brake amplitude and escalator technical parameters and uploading the data to the braking performance database;

B. Deep learning to establish an escalator braking force prediction model and an escalator braking force fault diagnosis model;

C. The braking performance database stores and reduces the data;

D. The reduced data is compared with the braking force prediction model and the braking force fault diagnosis model;

E. Comparison results View real-time data and give early warning through the braking force intelligent diagnosis platform.

The collection of the braking distance in the step A includes an encoder module and a braking signal module respectively connected to the processor; the processor wirelessly transmits the processed braking distance to the braking performance database;

The acquisition of the brake amplitude includes a vibration sensor connected to the processor, and the processor wirelessly transmits the processed brake amplitude data to the brake performance database.

The step B deep learning establishes an escalator braking force estimation model including two layers of RBM and one layer of BP; the two layers of RBM are fully connected, and the BP network is one-way connection;

The first RBM display layer is the braking force influencing factors, including escalator lifting height, nominal speed, nominal width, inclination angle, brake arm amplitude, maximum initial braking force, no-load up-stop distance, no-load down-stop distance , using the environment level, the input layer has 9 neurons;

The number of neurons in the second layer of RBM is 90;

The second hidden layer h2 is the input layer of the BP network, and finally the output power prediction result is controlled by the output layer of the BP network.

The escalator braking force fault diagnosis model is established, and useful fault features are directly extracted from the original fault data and the fault state is identified;

Clarify the type of escalator braking force failure;

Collect the braking distance data and brake vibration data of the escalator braking force fault type as input layer parameters;

A classifier is added to the top layer of the input layer, the preprocessed fault data is input into the DBM braking force early warning model, and the features of the fault diagnosis representative data are extracted from the model layer by layer;

The test set is input into the trained diagnostic model to identify the health of the device.

The types of escalator braking force failures include brake jamming, unsynchronized action of brake arms and insufficient braking torque.

A multi-sensor information processing for preprocessing data is also included between the braking performance database and the braking force intelligent diagnosis platform, and the multi-sensor information processing is a multi-dimensional time series of CPCA_DTW.

The beneficial effect of the present invention is that the deep belief network is used to estimate the braking force range of the escalator, and the useful fault features are directly extracted from the original data of the escalator failure and the fault state is identified, so the early warning and the abnormal braking force can be obtained at the first time. The diagnosis results of the braking force failure can timely check the hidden dangers of the escalator, realize intelligent management, and improve the operation efficiency of the escalator.

Description of drawings

Figure 1 is a schematic structural diagram of a restricted Boltzmann machine (RBM) model;

Figure 2 is a schematic diagram of the structure of a deep belief network (DBN) model;

Fig. 3 is the overall block diagram of the present invention;

Fig. 4 is a block diagram of an online detection device for escalator stopping distance;

Figure 5 is a block diagram of an escalator brake amplitude on-line detection device;

FIG. 6 is a schematic structural diagram of establishing an escalator braking force estimation model based on DBN;

Figure 7 is a schematic diagram of the process of fine-tuning the network weights through the BP algorithm;

8 is a schematic diagram of the process of establishing and training a fault diagnosis model based on DBN;

Fig. 9 is the data reduction process schematic diagram of the multidimensional time series based on CPCA_DTW;

Fig. 10 is a working flow chart of the intelligent diagnosis and monitoring platform for escalator braking force.

Detailed ways

There are many factors that affect the braking force of escalators. There is no specific mathematical model to explore the relationship between braking force and braking distance and other factors. Moreover, different models will have certain differences in different usage environments. Therefore, the braking force of each escalator Models are all unique. With the popularization of cutting-edge technology, artificial intelligence machine learning methods are more and more widely used in special equipment. The deep belief network (Deep Belief Network, DBN) of unsupervised learning in deep learning is derived from the research on the simulated brain nervous system of bionics. Realize the cognition and judgment of objects directly from the original data. The deep belief network theory can be applied to the estimation of the escalator braking force value and the diagnosis of the abnormal braking force. By directly starting from the original data, it can classify and identify the basic parameters of the escalator, the braking distance and the amplitude of the braking arm. The artificial feature extraction process reduces the human participation factor and enhances the accuracy of estimating the escalator braking force value and the intelligence of the abnormal braking force judgment.

As shown in Figure 1, the Restricted Boltzmann Machines (RBM) model is a kind of stochastic neural network model with two layers of explicit and hidden layers, which can be understood as a special Markov The random field has the characteristic that the neurons inside the explicit layer and the hidden layer are not connected, while the neurons between the explicit layer and the hidden layer are all connected.

In RBM, v represents the neurons in the explicit layer, h represents the neurons in the hidden layer, W represents the weight of the connection strength between any two adjacent neurons, a represents the self-bias of the neurons in the explicit layer, and b represents the neurons in the hidden layer. Meta self-bias. The energy of an RBM can be expressed as:

The probability that the hidden layer neuron hj is activated is:

P(h _j |v)=σ(b _i +∑ _i W _ij v _i ) (2)

Since the hidden layer and the visible layer are fully connected, the visible layer can also be activated by the hidden layer;

P(v _i |h)=σ(a _i +∑ _j W _ij h _j ) (3)

为sigmoid函数，此时函数值域为[0,1]用来描述神经元被激活的概率。同一层神经元无连接，故神经元满足概率面密度独立性：in,

It is a sigmoid function, and the function value range is [0,1] to describe the probability of the neuron being activated. The neurons in the same layer are not connected, so the neurons satisfy the probability surface density independence:

When a certain piece of data x is assigned to the display layer, RBM can calculate the probability P(h _j |x) of each hidden layer neuron being activated according to formula (2), taking μ∈[0,1] as the threshold, when When P(h _j |x)≥μ, h _j =1, when P(h _j |x)<μ, h _j =0, which can be obtained whether each hidden layer nerve is activated. An RBM model can be determined according to three parameters a, b and W.

As shown in Figure 2, the Deep Belief Network (DBN) is composed of several layers of neurons, and its original component is a restricted Boltzmann machine (RBM). DBN, the hidden layer of the previous RBM is the visible layer of the next RBM, as shown in Figure 2. Therefore, the deep belief network can also be interpreted as a Bayesian probability generation model composed of multiple layers of random latent variables. The first layer of the display layer can be input data. In the training phase, the relevant information is extracted from the display layer through Gibbs sampling and mapped to the hidden layer. The input data is constructed, the mapping and reconstruction process between the visible layer and the hidden layer is repeatedly performed, and the values of the weights a, b and W are continuously updated in each reconstruction process.

The present invention is described in further detail below in conjunction with the accompanying drawings and specific embodiments:

As shown in FIG. 3, an intelligent diagnosis system for escalator braking force based on deep belief network of the present invention includes steps:

A. Collection of braking distance, brake amplitude and escalator technical parameters and uploading the data to the braking performance database;

As shown in Figure 4, the collection of braking distance includes an encoder module and a braking signal module respectively connected to the processor; the encoder module is installed at the entrance of the return section of the handrail and is connected to the processor through a cable, and the braking signal module Installed on the brake contactor of the control cabinet, it is connected to the processor through a cable, and the processor wirelessly transmits the braking distance obtained from the processing to the braking performance database.

As shown in Figure 5, the acquisition of the brake amplitude includes a vibration sensor connected to the processor; the vibration sensor is installed on the two brake arms, and each time the escalator brake is opened and closed, the vibration sensor will be triggered to work, The vibration sensor is connected to the processor, and the processor wirelessly transmits the processed brake amplitude data to the brake performance database.

B. Deep learning establishes an escalator braking force estimation model and an escalator braking force fault diagnosis model, which can estimate the braking force range of escalators of different models and different operating environments;

As shown in Figure 6, the deep learning establishment of the escalator braking force estimation model includes two layers of RBM and one layer of BP; the two layers of RBM are fully connected, and the BP network is a one-way connection;

The first RBM display layer is the braking force influencing factors, including escalator lifting height, nominal speed, nominal width, inclination angle, brake arm amplitude, maximum initial braking force, no-load up-stop distance, no-load down-stop distance , using the environment level, the input layer has 9 neurons;

There are 90 neurons in the second layer of RBM; the second layer of hidden layer h2 is the input layer of the BP network, and finally the power prediction results are output through the output layer of the BP network.

As shown in Figure 7, the training process of the deep belief network is as follows: the basic technical parameters of the equipment are directly used as input layer parameters; as input layer parameters. Determine the learning rate, learning direction, the maximum number of iterations of training, the maximum number of iterations of fine-tuning, the number of samples and the training batch, set the threshold function of the neural network algorithm (Neural Network) as the sigmoid function, and calculate the probability of h1 through v1, Gibbs sampling: h ₁ ~P(h|v ₁ ), and then reconstruct the display layer v1 through h1, that is, use the hidden layer to infer the probability distribution P(v|h ₁ ) of the display layer, and after satisfying the accuracy requirements, The first layer of RBM training is completed. The training of the second layer of RBM starts, and h1 is used as the input layer v2 of the second layer of RBM. Continue the above RBM training method, and calculate the probability of v2 through h1. Gibbs sampling: v ₂ ～P(v|h ₁ ) Pass v2 calculates the probability of h2 Gibbs sampling h ₂ ～P(h|v ₂ ) completes the training of the second-layer RBM after satisfying the requirements. The training result of h2 is supervised training for the input factors of the BP network, fine-tuning the network weights according to the BP error back propagation algorithm, and updating the weight content:

W←W+λ(P(h ₁ |v ₁ )v ₁ -P(h ₂ |v ₂ )v ₂ )

a←a+λ(v ₁ -v ₂ )

b←b+λ(v ₁ -v ₂ )

In this way, the model converges to the local optimum point, and finally the estimated braking force value is obtained, and the entire estimated model network is completed.

In the aspect of fault diagnosis, deep belief network can effectively overcome the shortcomings of the traditional shallow learning method's insufficient feature expression ability when faced with complex and huge data, and can directly extract useful fault features from the original fault data and identify the fault state. .

As shown in Figure 8, the establishment and training of the escalator fault diagnosis model based on DBN:

1) First of all, it is necessary to clarify the types of escalator braking force failures, which can be basically divided into failures such as brake jamming, asynchronous brake arm action, and insufficient braking torque.

2) These states are artificially simulated, and the braking distance data and brake vibration data in the corresponding states are collected as input layer parameters.

3) Add a classifier on the top layer of the input layer, and then input the preprocessed fault data into the DBM braking force early warning model, use the fault samples to train the model, and extract the features of the fault diagnosis representative data layer by layer from the model.

4) Use the network to perform supervised fine-tuning of the relevant parameters of the DBN. After the fine-tuning is completed, the test set is input into the trained diagnostic model to identify the health status of the equipment.

C. The braking performance database stores and reduces the data;

The parameters of the braking distance signal and the vibration signal of the brake arm are collected, and the two signals are collected by a displacement sensor and a vibration sensor respectively. After the signals are obtained, the data can be preprocessed as the input layer parameters for fault diagnosis. Data preprocessing here is to integrate and reduce the data of two different sensor signals, so as to obtain an information processing method of multi-sensor representation data.

Fault diagnosis methods based on multi-sensor monitoring information processing can be roughly divided into four categories: data layer fusion, feature layer fusion, decision layer fusion and multi-dimensional time series classification. The present invention adopts a multi-dimensional time series classification method.

Multidimensional time series classification methods mainly include distance-based methods and feature-based methods. This paper adopts two methods to combine multi-dimensional time series based on CPCA_DTW, namely based on common principal component analysis (Common PCA, CPCA) and dynamic time warping distance (Dynamic Time Warping, DTW). CPCA reduces the dimension of the sequence on the basis of ensuring the extracted features, and DTW distance can find the minimum distance measure between two time series through a certain optimization method.

D. The reduced data is compared with the braking force prediction model and the braking force fault diagnosis model;

As shown in Figure 9, the specific solution is to first use CPCA to reduce the dimension of the signal samples, calculate the average covariance matrix and eigenvector, and get the reduced sequence after sorting the eigenvalues; then use DTW distance to calculate the reduced sequence. The distance between the two vectors is recorded, the distance matrix is recorded, and the reduced sequence is stacked to form a one-dimensional vector; finally, the matrix and the one-dimensional vector are combined to form a one-dimensional data vector that can be input into the DBN model.

E. Comparison results View real-time data and give early warning through the braking force intelligent diagnosis platform.

As shown in Figure 10, through the escalator braking force prediction model based on the deep belief network, we can infer the range of the braking force of an escalator; through the escalator fault diagnosis model based on the deep belief network, we realize the braking force correlation Troubleshooting. On this basis, the present invention builds a braking force intelligent diagnosis and monitoring platform to realize information-based supervision, and integrates real-time monitoring, data statistics, intelligent early warning and fault diagnosis.

Complete the acquisition equipment arrangement on the escalator site, including the braking distance detection sensor and the brake vibration monitoring sensor. Establish and train the braking force prediction model and the braking force fault diagnosis model in the cloud system. When the escalator is in normal use, the collection device extracts the data related to the braking force in each normal stop or emergency braking condition, that is, the brake amplitude, the up/down braking distance, etc. The data is stored in the IoT cloud server. and reduce. The reduced data is compared with the prediction model and the fault diagnosis model to realize monitoring and analysis. When the braking force declines to a sub-healthy state, it enters the early warning range, and sends an early warning signal to the Internet of Things cloud server to remind the user and maintenance unit to take corresponding measures. measure. When a suspected fault occurs, the system extracts fault features, compares them with the fault diagnosis model, diagnoses the braking force fault in real time, and comprehensively analyzes the braking force state through the estimated value given by the prediction model and the fault category diagnosed by the fault diagnosis model. . After data acquisition, users can view real-time data, historical data and diagnostic results through a browser.

The user unit of the present invention can grasp the braking force status of the escalator in real time through the intelligent diagnosis platform, obtain the early warning of the abnormal braking force and the diagnosis result of the braking force failure at the first time, timely check the hidden dangers of the escalator, realize the intelligent management, and improve the comprehensive service. Level.

The maintenance unit can timely find out the cause of the braking force failure through the intelligent diagnosis platform, and grasp the trend of braking force decrease and braking distance increase, carry out targeted troubleshooting, and carry out preventive maintenance to reduce escalator failures rate and accident rate, and improve the service quality of maintenance units.

Through the intelligent diagnosis platform, the manufacturing unit can carry out long-term information management of the escalator of the manufacturer, explore the potential factors that induce braking force-related failures, provide data support for optimizing the design process of the braking system, and improve product quality and market competitiveness.

The inspection department can obtain the historical data of the escalator braking force failure and the long-term change trend of the braking force through the intelligent diagnosis platform, so as to grasp the braking force status and maintenance quality of the inspected escalator.

It is worth noting that the protection scope of the present invention is not limited to the above-mentioned specific examples. According to the basic technical concept of the present invention, the purpose of the present invention can also be achieved by using basically the same structure, as long as those of ordinary skill in the art do not need to be creative. Work, the embodiments that can be thought of, all belong to the protection scope of the present invention.

Claims (6)

1. a kind of intelligent diagnosis system of escalator braking force based on deep belief network, is characterized in that, comprises step, A. Collect braking distance, brake amplitude and technical parameters of escalator and upload data to braking performance database; b. Deep learning establishes an escalator braking force estimation model and an escalator braking force fault diagnosis model; c. The braking performance database stores and reduces the data; D. The reduced data is compared with the braking force prediction model and the braking force fault diagnosis model; E. The comparison results are used to view real-time data and give early warning through the braking force intelligent diagnosis platform. 2. the escalator braking force intelligent diagnosis system based on deep belief network according to claim 1, is characterized in that, The collection of the braking distance in the step A includes an encoder module and a braking signal module respectively connected to the processor; the processor wirelessly transmits the processed braking distance to the braking performance database; The acquisition of the brake amplitude includes a vibration sensor connected to the processor, and the processor wirelessly transmits the processed brake amplitude data to the brake performance database. 3. the escalator braking force intelligent diagnosis system based on deep belief network according to claim 1, is characterized in that, Described step B deep learning establishes escalator braking force estimation model and comprises two-layer RBM and one-layer BP; Described two-layer RBM is full connection, and BP network is one-way connection; The first RBM display layer is the braking force influencing factors, including escalator lifting height, nominal speed, nominal width, inclination angle, brake arm amplitude, maximum initial braking force, no-load up braking distance, no-load down braking distance , using the environment level, the input layer has 9 neurons; The number of neurons in the second layer of RBM is 90; The second hidden layer h2 is the input layer of the BP network, and finally the output power prediction result is controlled by the output layer of the BP network. 4. the escalator braking force intelligent diagnosis system based on deep belief network according to claim 1, is characterized in that, The escalator braking force fault diagnosis model is established, and useful fault features are directly extracted from the original fault data and the fault state is identified; Clarify the type of escalator braking force failure; Collect the braking distance data and brake vibration data of the escalator braking force fault type as input layer parameters; A classifier is added to the top layer of the input layer, the preprocessed fault data is input into the DBM braking force early warning model, and the features of the fault diagnosis representative data are extracted from the model layer by layer; The test set is input into the trained diagnostic model to identify the health of the device. 5 . The intelligent diagnosis system for escalator braking force based on deep belief network according to claim 4 , wherein the escalator braking force fault types include brake jamming, unsynchronized action of brake arm and insufficient braking torque. 6 . . 6. the escalator braking force intelligent diagnosis system based on deep belief network according to claim 1, is characterized in that, A multi-sensor information processing for preprocessing data is also included between the braking performance database and the braking force intelligent diagnosis platform, and the multi-sensor information processing is a multi-dimensional time series of CPCA_DTW.

Images