Elevator health prediction method and system based on deep learning

Abstract

The invention discloses an elevator health prediction method and system based on deep learning, and relates to the technical field of health prediction of special equipment. Comprising the steps of obtaining operation state data of a to-be-evaluated elevator in multiple operation processes; collecting the fault operation state data of the to-be-evaluated elevator to obtain a training set, and collecting the normal operation state data to obtain a verification set; constructing a health life prediction network; inputting the training set into a health life prediction network, and updating network weight parameters through a loss function to obtain a trained health life prediction model; and inputting the verification set of the to-be-evaluated elevator into the trained health life prediction model, evaluating the health life prediction result of the elevator, and outputting the health life condition level of the elevator. Multi-dimensional health inspection on the elevator is facilitated, the health condition of the elevator is scientifically analyzed, potential safety hazards of the elevator and safety parts are pre-judged, life health prediction of the elevator in the whole life cycle is achieved, and elevator safety accidents are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of special equipment health prediction, and in particular to an elevator health prediction method and system based on deep learning. Background Art

[0002] As a common special equipment in daily life, elevators bring convenience to people's travel, but their operating risks are also safety hazards that cannot be ignored. The main causes of elevator accidents are inadequate safety management and maintenance, illegal operations or improper operation, failure of safety components or protection devices, failure to inspect in time, improper emergency rescue and other reasons. If the health status and risks of the elevator can be evaluated and predicted before an accident occurs, unnecessary casualties will be reduced.

[0003] In the prior art, the mainstream elevator detection is to compare the information collected by the sensor acquisition end with the data of the normal operation of the main control end to determine the elevator fault situation. This judgment method requires the comparison of multiple different parameters at the same time. There are many parameters such as speed, temperature, vibration amplitude, etc. during the operation of the elevator. Comparing these parameters at the same time will cause a large CPU interrupt load and high system reliability requirements. In addition, it is difficult to accurately determine the wear and stability of the components represented by each operating parameter through simple data comparison. At the same time, it is also difficult to reasonably and effectively judge the overall operation health of the elevator, and it is impossible to provide maintenance personnel with more elevator operation health information. In order to achieve the reasonable allocation of elevator maintenance resources, maintenance personnel need to first evaluate the health status of the elevator and formulate corresponding maintenance plans based on the health status of the elevator.

[0004] Therefore, it is an urgent problem for technical personnel in this field to propose an elevator health prediction method and system based on deep learning, so as to conduct multi-dimensional health checks on elevators and make health predictions for the entire life cycle of the elevator. Summary of the invention

[0005] In view of this, the present invention provides an elevator health prediction method and system based on deep learning, which is conducive to multi-dimensional health checks on elevators, scientific analysis of the health status of elevators and prediction of safety hazards of elevators and safety components, realizing life health prediction of elevators throughout their life cycle, and reducing the occurrence of elevator safety accidents.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for predicting elevator health based on deep learning, comprising the following steps:

[0008] S1. Obtaining fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations;

[0009] S2. The fault operation status data set of the elevator to be evaluated is used to obtain a training set, and the normal operation status data set is used to obtain a verification set;

[0010] S3. Construct a healthy life expectancy prediction network;

[0011] S4. Input the training set into the healthy life span prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life span prediction model after several trainings;

[0012] S5. Input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction result, and output the elevator healthy life status level.

[0013] In the above method, optionally, during one operation of the elevator to be evaluated in S1, different operating parameters of the elevator are collected once at a fixed interval, so that the elevator to be evaluated runs multiple times and the operating data of each time is obtained respectively; if the elevator is in a normal operating state, the data of the elevator operation is marked, otherwise it is not marked; and the fault operating state data and the normal operating state data of the elevator to be evaluated during multiple operations are obtained.

[0014] The above method may optionally further include preprocessing the fault operation status data and the normal operation status data obtained in S1 by removing empty data, removing discrete data, and normalizing.

[0015] In the above method, optionally, the healthy life span prediction network constructed in S3 includes Conv+BN+ReLU layer, multiple time-frequency attention residual blocks, global average pooling layer, and fully connected layer.

[0016] In the above method, optionally, the Conv+BN+ReLU layer is composed of a convolutional layer, a BN layer, and a ReLU layer;

[0017] The time-frequency attention residual block consists of two convolutional layers, two BN layers, two ReLU layers and a time-frequency attention unit;

[0018] Convolutional layer, used to extract data features;

[0019] The BN layer is used to normalize the input feature values ​​to a mean value of 0 and a standard deviation of 1, reducing the distribution differences of features between different batches;

[0020] The ReLU layer is used to realize the nonlinear transformation from input data to output data.

[0021] In the above method, optionally, in S4, the training set is input into the healthy life prediction network, and the network weight parameters are updated through the loss function. After several trainings, the specific content of the trained healthy life prediction model is obtained as follows:

[0022] Input the training set, initialize the parameters of the healthy life expectancy prediction model, perform feature extraction, calculate the outputs of the Conv+BN+ReLU layer, the time-frequency attention residual module, and the global average pooling layer, update the model parameters, output the optimized model parameters, and obtain the trained healthy life expectancy prediction model.

[0023] The above method is optional. The elevator health life status level output in S5 is divided into five levels: "level 1 healthy life status", "level 2 healthy life status", "level 3 healthy life status", "level 4 healthy life status" and "level 5 healthy life status", and a maintenance report and inspection order is generated according to the elevator health life level status.

[0024] An elevator health prediction system based on deep learning, applying any one of the above elevator health prediction methods based on deep learning, comprising: a data acquisition module, a data partitioning module, a network construction module, a model training module and a health prediction module connected in sequence;

[0025] A data acquisition module is used to acquire fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations;

[0026] A data partitioning module is used to obtain a training set from the elevator fault operation status data set to be evaluated, and obtain a verification set from the normal operation status data set;

[0027] Network construction module, used to build a healthy lifespan prediction network;

[0028] The model training module is used to input the training set into the healthy life prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life prediction model after several trainings;

[0029] The health prediction module is used to input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction results, and output the elevator healthy life status level.

[0030] It can be seen from the above technical solution that compared with the prior art, the present invention provides an elevator health prediction method and system based on deep learning, which has the following beneficial effects: it is conducive to multi-dimensional health checks on elevators, scientific analysis of the health status of elevators and prediction of safety hazards of elevators and safety components, and realization of life health prediction of elevators throughout their life cycle, thereby reducing the occurrence of elevator safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] 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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0032] Figure 1 A flow chart of an elevator health prediction method based on deep learning provided by the present invention;

[0033] Figure 2 A structural block diagram of an elevator health prediction system based on deep learning provided by the present invention. DETAILED DESCRIPTION

[0034] 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.

[0035] Reference Figure 1 As shown, the present invention discloses a method for predicting elevator health based on deep learning, comprising the following steps:

[0036] S1. Obtaining fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations;

[0037] S2. The fault operation status data set of the elevator to be evaluated is used to obtain a training set, and the normal operation status data set is used to obtain a verification set;

[0038] S3. Construct a healthy life expectancy prediction network;

[0039] S4. Input the training set into the healthy life span prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life span prediction model after several trainings;

[0040] S5. Input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction result, and output the elevator healthy life status level.

[0041] Furthermore, during one operation of the elevator to be evaluated in S1, different operating parameters of the elevator are collected at a fixed interval, so that the elevator to be evaluated runs multiple times and the operating data of each time is obtained respectively; if the elevator is in a normal operating state, the data of the elevator operation is marked, otherwise it is not marked; the fault operating state data and the normal operating state data of the elevator to be evaluated during multiple operations are obtained.

[0042] Specifically, the operating status data during the operation of the elevator includes the operating status data of the traction machine, the operating status data of the wire rope, the operating status data of the control cabinet, the operating status data of the control cabinet, the operating status data of the intermediate relay and the phase sequence relay, the operating status data of the guide rail, the operating status data of the circuit breaker, the operating status data of the accompanying cable, the operating status data of the speed limiter, the operating status data of the car, the operating status data of the counterweight block and the operating status data of the shaft parts.

[0043] The operating parameters of the elevator during operation include the deviation between the operating speed and the rated speed, rated load, leveling accuracy, machine room noise, noise in the car during operation, noise during the opening and closing process, operating temperature, and operating current and voltage.

[0044] Furthermore, it also includes preprocessing of removing empty data, removing discrete data and normalizing the fault operation status data and normal operation status data obtained in S1.

[0045] Specifically, removing empty data includes: removing data with negative values, Null or None; removing discrete data includes: removing data that is obviously unreasonable, such as the number of trapped people exceeding 100 times in a single day; normalization processing can process all data into the range of [0,1];

[0046] Using the minimum and maximum normalization method, the normalization formula is:

[0047]

[0048] Among them, x i Represents input data; x min and x max Respectively represent x i The minimum and maximum values ​​of X N Represents the normalized output.

[0049] Furthermore, the healthy life span prediction network constructed in S3 includes Conv+BN+ReLU layers, multiple time-frequency attention residual blocks, global average pooling layers, and fully connected layers.

[0050] Furthermore, the Conv+BN+ReLU layer is composed of a convolutional layer, a BN layer, and a ReLU layer;

[0051] The time-frequency attention residual block consists of two convolutional layers, two BN layers, two ReLU layers and a time-frequency attention unit;

[0052] Convolutional layer, used to extract data features;

[0053] The BN layer is used to normalize the input feature values ​​to a mean value of 0 and a standard deviation of 1, reducing the distribution differences of features between different batches;

[0054] The ReLU layer is used to realize the nonlinear transformation from input data to output data.

[0055] Specifically, Conv(64,7×7,s=2) means that the number of output channels of the convolution layer is 64, the size of the convolution kernel is 7×7, and the number of convolution kernel movement strides s is equal to 2; the healthy life prediction model takes the original data as input and the specific state or fault category as output, which is an end-to-end deep learning model; the original data is processed by STFT to obtain the time-frequency coefficient matrix, which will pass through a series of convolution layers, BN layers, ReLU nonlinear activation function layers, time-frequency attention mechanism units, global average pooling layers, fully connected layers, etc., a total of 17 layers of convolution operations and 1 layer of fully connected layer operations are required; the number of output neurons of the fully connected layer is the number of categories of state monitoring or fault diagnosis; in the first convolution layer, a wide convolution kernel of 7×7 is used, which helps the model suppress noise interference, the number of movement strides of the first convolution layer is 2, and the number of movement strides of other convolution layers is 1, which helps to retain more important feature information.

[0056] Specifically, the output of the fully connected layer needs to go through the softmax activation function to convert the output value of the neuron to between 0 and 1, and the sum of all values ​​is 1. The softmax activation function can be expressed as:

[0057]

[0058] Where N represents the number of categories; x i represents the output of the fully connected layer; p j Represents x j Output after softmax.

[0059] Furthermore, in S4, the training set is input into the healthy life prediction network, and the network weight parameters are updated through the loss function. After several trainings, the specific content of the trained healthy life prediction model is obtained as follows:

[0060] Input the training set, initialize the parameters of the healthy life expectancy prediction model, perform feature extraction, calculate the outputs of the Conv+BN+ReLU layer, the time-frequency attention residual module, and the global average pooling layer, update the model parameters, output the optimized model parameters, and obtain the trained healthy life expectancy prediction model.

[0061] Furthermore, the elevator health life status level output in S5 is divided into five levels: "level 1 healthy life status", "level 2 healthy life status", "level 3 healthy life status", "level 4 healthy life status" and "level 5 healthy life status", and a maintenance report and inspection order is generated according to the elevator health life level status.

[0062] Specifically, the healthy life status of elevators is graded into five levels: "level one healthy life status", "level two healthy life status", "level three healthy life status", "level four healthy life status" and "level five healthy life status".

[0063] Level 1 healthy life status: It is a healthy operating status, records and analyzes data, provides a reference for preventive maintenance, and does not require early warning.

[0064] Level 2 healthy life status: It is a low-risk operating state, which prompts the staff to pay attention to the status of the component and recommends regular inspection or maintenance.

[0065] Level 3 healthy life status: It is a medium-risk operating state. At this time, the staff will be notified and reminded to carry out maintenance in time.

[0066] Level 4 healthy life status: It is a high-risk operating state. At this time, notify the staff to carry out immediate maintenance and stop working.

[0067] Level 5 Healthy Life Status: This is an extremely high-risk operating state. The elevator must be stopped immediately and repaired or replaced immediately.

[0068] In addition, a maintenance report and inspection sheet are generated based on the five levels of healthy life status of the elevator, recording the details of each treatment and providing a basis for subsequent maintenance management and data analysis.

[0069] In a specific embodiment, during the operation of the elevator, different operating parameters of the elevator are collected at fixed intervals, and the elevator to be tested is run multiple times to obtain the operating data of each time; if the elevator is in normal operation, the data of the elevator operation is marked, otherwise it is not marked; the fault operating state data and normal operating state data of the elevator to be evaluated during multiple operations are obtained. The fault operating state data and normal operating state data of the elevator are recorded for 500 times, and the healthy life prediction network is trained using the elevator fault operating state data to reduce the loss function value and update the network weight. After continuous iterative learning, a healthy life prediction model is obtained.

[0070] The specific network training process can be:

[0071] The training set is input into the healthy life span prediction network for K iterations of training;

[0072] Calculate the minimum loss function of the nth iteration training result, and adjust the learning rate of the n+1th iteration training according to the minimum loss function;

[0073] Update the model parameters by measuring the K-times iterative training results through the loss function and output the optimized model parameters;

[0074] A healthy lifespan prediction model is generated based on the training results of K iterations of training.

[0075] In the embodiment of the present invention, during the above model training process, the detailed training parameters can be set as:

[0076] (1) Epochs = 60, which means that the training will be iterated 60 times;

[0077] (2) BatchSize = 43, indicating the batch size of samples input to the health life prediction network at one time;

[0078] (3) Loss Function = "categorical_crossentropy", indicating the use of loss function;

[0079] (4) Optimizer = "Adadelta", which means to constrain the learning rate of the healthy life prediction model training.

[0080] Specifically, by minimizing the classification cross entropy and constraining the learning rate of the neural network model training, the predicted value can be close to the true value.

[0081] Specifically, by increasing the amount of data in the training set and the test set, the error can be reduced and the overfitting phenomenon in the training of the healthy life span prediction model can be avoided.

[0082] and Figure 1 Corresponding to the method described above, the embodiment of the present invention also provides an elevator health prediction system based on deep learning, which is used to Figure 1 The specific implementation of the method is shown in the following diagram: Figure 2 As shown, it specifically includes: a data acquisition module, a data partitioning module, a network construction module, a model training module and a health prediction module connected in sequence;

[0083] A data acquisition module is used to acquire fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations;

[0084] A data partitioning module is used to obtain a training set from the elevator fault operation status data set to be evaluated, and obtain a verification set from the normal operation status data set;

[0085] Network construction module, used to build a healthy lifespan prediction network;

[0086] The model training module is used to input the training set into the healthy life prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life prediction model after several trainings;

[0087] The health prediction module is used to input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction results, and output the elevator healthy life status level.

[0088] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device 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 method part.

[0089] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting elevator health based on deep learning, characterized in that: The following steps are involved: S1. Obtaining fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations; S2. The fault operation status data set of the elevator to be evaluated is used to obtain a training set, and the normal operation status data set is used to obtain a verification set; S3. Construct a healthy life expectancy prediction network; S4. Input the training set into the healthy life span prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life span prediction model after several trainings; S5. Input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction result, and output the elevator healthy life status level.

2. The elevator health prediction method based on deep learning according to claim 1 is characterized in that: In S1, during the operation of the elevator to be evaluated, different operating parameters of the elevator are collected at a fixed interval, so that the elevator to be tested runs multiple times and the operating data of each time is obtained; If the elevator is in normal operation, the data of the elevator operation will be marked, otherwise it will not be marked; Obtain fault operation status data and normal operation status data during multiple operations of the elevator to be evaluated.

3. The elevator health prediction method based on deep learning according to claim 1 is characterized in that: It also includes preprocessing of removing empty data, removing discrete data and normalizing the fault operation status data and normal operation status data obtained in S1.

4. The elevator health prediction method based on deep learning according to claim 1, characterized in that: The healthy life span prediction network constructed in S3 includes Conv+BN+ReLU layers, multiple time-frequency attention residual blocks, global average pooling layers, and fully connected layers.

5. The elevator health prediction method based on deep learning according to claim 4 is characterized in that: The Conv+BN+ReLU layer consists of a convolutional layer, a BN layer, and a ReLU layer; The time-frequency attention residual block consists of two convolutional layers, two BN layers, two ReLU layers and a time-frequency attention unit; Convolutional layer, used to extract data features; The BN layer is used to normalize the input feature values ​​to a mean value of 0 and a standard deviation of 1, reducing the distribution differences of features between different batches; The ReLU layer is used to realize the nonlinear transformation from input data to output data.

6. The elevator health prediction method based on deep learning according to claim 1, characterized in that: In S4, the training set is input into the healthy life prediction network, and the network weight parameters are updated through the loss function. After several trainings, the specific content of the trained healthy life prediction model is as follows: Input the training set, initialize the parameters of the healthy life expectancy prediction model, perform feature extraction, calculate the outputs of the Conv+BN+ReLU layer, the time-frequency attention residual module, and the global average pooling layer, update the model parameters, output the optimized model parameters, and obtain the trained healthy life expectancy prediction model.

7. The elevator health prediction method based on deep learning according to claim 1, characterized in that: The elevator health life status level output in S5 is divided into five levels: "level 1 health life status", "level 2 health life status", "level 3 health life status", "level 4 health life status" and "level 5 health life status", and a maintenance report and inspection order is generated according to the elevator health life level status.

8. An elevator health prediction system based on deep learning, characterized in that: An elevator health prediction method based on deep learning according to any one of claims 1 to 7 is applied, comprising: a data acquisition module, a data partitioning module, a network construction module, a model training module and a health prediction module connected in sequence; A data acquisition module is used to acquire fault operation status data and normal operation status data of the elevator to be evaluated during multiple operations; A data partitioning module is used to obtain a training set from the elevator fault operation status data set to be evaluated, and obtain a verification set from the normal operation status data set; Network construction module, used to build a healthy lifespan prediction network; The model training module is used to input the training set into the healthy life prediction network, update the network weight parameters through the loss function, and obtain the trained healthy life prediction model after several trainings; The health prediction module is used to input the verification set of the elevator to be evaluated into the trained healthy life prediction model, evaluate the elevator healthy life prediction results, and output the elevator healthy life status level.

Images