Traction sheave wear fault diagnosis method based on digital twinning

Abstract

The invention discloses a traction sheave wear fault diagnosis method based on digital twinning, and relates to the field of elevator wear, and the method comprises the steps: carrying out the rigid-flexible coupling dynamics simulation of a three-dimensional model of a traction elevator, obtaining the acceleration of a lift car and the tension of a steel wire rope of each wheel groove at a lift car side and a counterweight side in a wear process, and obtaining a data set; using an IPOA algorithm to optimize the hyper-parameter of the LSTM; the IPOA algorithm is a pelican optimization algorithm in which a population is initialized by using a method of generating a pseudo-random number by using a Halton sequence and individual positions are updated by using a self-adaptive nonlinear weight factor based on a sine function; training and testing the optimized CNN-LSTM neural network according to the data set to obtain a fault diagnosis model; according to the method, the high-precision, real-time and automatic traction sheave abrasion fault diagnosis effect can be achieved, and the problem of abrasion fault diagnosis of the elevator traction sheave is solved.

Description

Technical Field

[0001] The present application relates to the field of elevator wear, and in particular to a traction wheel wear fault diagnosis method based on digital twin. Background Art

[0002] As a key component in the elevator system, the elevator traction sheave directly affects the safety and operating efficiency of the elevator. The traction sheave realizes the rise and fall of the elevator through the friction between the traction sheave and the wire rope, and its wear condition has an important impact on the performance of the entire elevator system. How to scientifically judge whether the traction sheave has failed is a difficult problem that needs to be solved urgently. Studying the influence of traction sheave wear on the equivalent friction coefficient is an effective solution to scientifically judge the failure of the traction sheave.

[0003] Most existing detection methods rely on manual operation, through manual observation of the wear of the traction sheave surface. This method is simple, but has a low level of automation, low efficiency, is only suitable for preliminary inspections, and has the problem of inconsistent test results. This method is not only time-consuming and labor-intensive, but also difficult to adapt to the needs of modern elevator management for efficient and accurate detection.

[0004] Although some researchers have used image processing methods to detect wear, camera detection has high requirements for lighting conditions. Too strong or too weak light will affect the image quality, which in turn affects the detection accuracy. In many cases, detection is required when the equipment is shut down to ensure that a clear image is captured. The wear condition of the traction wheel cannot be monitored in real time, which not only affects the normal operation of the elevator, but also easily causes the wear problem to deteriorate if it is not discovered in time, increasing safety hazards. The implementation of real-time online detection is difficult, especially when running at high speeds, it is difficult to obtain clear and stable images.

[0005] Traditional inspection methods lack in-depth analysis and utilization of inspection data, and cannot provide detailed wear trend analysis and prediction. This makes it difficult to achieve early prevention and planned maintenance, and can only rely on periodic inspections and post-processing.

[0006] Therefore, a high-precision, real-time and automated traction sheave wear fault diagnosis method is needed to solve the problem of elevator traction sheave wear fault diagnosis. Summary of the invention

[0007] The purpose of this application is to provide a traction wheel wear fault diagnosis method based on digital twin, which can achieve high-precision, real-time and automated traction wheel wear fault diagnosis effect, and solve the problem of elevator traction wheel wear fault diagnosis.

[0008] To achieve the above objectives, this application provides the following solutions:

[0009] In a first aspect, the present application provides a traction sheave wear fault diagnosis method based on digital twin, comprising:

[0010] Under each equivalent friction coefficient between the wire rope and the traction wheel, the rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator to obtain the sensor data corresponding to each wear amount during the normal wear process and the abnormal wear process under each equivalent friction coefficient, and obtain a data set; the sensor data during the normal wear process includes the car acceleration during the normal wear process, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the sensor data during the abnormal wear process includes the single-groove wear abnormal data and the synchronous wear abnormal data; the single-groove wear abnormal data includes: the car acceleration during the abnormal wear process, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the synchronous wear abnormal data includes the car acceleration during the abnormal wear process, the wire rope tension of the target wheel groove on the car side, and the wire rope tension of the target wheel groove on the counterweight side;

[0011] The IPOA algorithm is used to optimize the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network to obtain the optimized CNN-LSTM neural network; the IPOA algorithm is a Pelican optimization algorithm that uses a Halton sequence to generate pseudo-random numbers to initialize the population and uses an adaptive nonlinear weight factor based on a sine function to update the individual position; the population includes multiple individuals, and one individual includes all the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network;

[0012] With car acceleration, wire rope tension of car side wheel groove and counterweight side wheel groove as input and wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained and tested according to the data set to obtain the fault diagnosis model; the wear fault diagnosis result is normal wear, single groove wear abnormality or synchronous wear abnormality;

[0013] The car acceleration of the traction sheave to be diagnosed at the current moment, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are input into the fault diagnosis model to obtain the wear fault diagnosis result of the traction sheave to be diagnosed.

[0014] Optionally, a rigid-flexible coupling dynamics simulation is performed on the three-dimensional model of the traction elevator under each equivalent friction coefficient between the wire rope and the traction wheel, and sensor data corresponding to each wear amount during normal wear and abnormal wear under each equivalent friction coefficient is obtained to obtain a data set, specifically including:

[0015] Under any equivalent friction coefficient between the wire rope and the traction wheel, a rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration corresponding to each wear amount, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are obtained in the process of the groove angle wearing from the initial angle to the first preset angle at a first preset wear amount interval, and the sensor data corresponding to each wear amount in the normal wear process under the equivalent friction coefficient are obtained;

[0016] A rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side corresponding to each wear amount in the process of the groove angle wearing from the first preset angle to the second preset angle are obtained at a second preset wear interval to obtain sensor data corresponding to each wear amount in the abnormal wear process under the equivalent friction coefficient; the first preset angle is greater than the second preset angle.

[0017] Optionally, the first preset angle is 25 degrees, and the second preset angle is 0 degrees.

[0018] Optionally, the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove are used as inputs, and the wear fault diagnosis result is used as output. The optimized CNN-LSTM neural network is trained and tested according to the data set to obtain a fault diagnosis model, which specifically includes:

[0019] Dividing the data set into a training set and a test set;

[0020] Taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as inputs and the wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained according to the training set to obtain the trained CNN-LSTM neural network;

[0021] The trained CNN-LSTM neural network is tested using the test set to obtain the test results;

[0022] If the test result does not meet the requirements, the step of taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as inputs and the wear fault diagnosis result as output, and training the optimized CNN-LSTM neural network according to the training set to obtain the trained CNN-LSTM neural network is returned;

[0023] If the test results meet the requirements, the fault diagnosis model is obtained.

[0024] Optionally, the CNN-LSTM neural network includes a convolutional neural network and a long short-term memory neural network connected in sequence.

[0025] Optionally, an adaptive nonlinear weight factor based on a sine function is given by:

[0026] w=w max -(w max -w min )·sin((t / T) m ·(π / 2)), where w represents the adaptive nonlinear weight factor, w max Indicates the maximum inertia weight, w min It represents the minimum value of inertia weight, m ​​is a real number and has no practical meaning, t represents the number of loops currently executed by the IPOA algorithm, T represents the maximum number of loops of the IPOA algorithm, and π represents the ratio of pi.

[0027] According to the specific embodiments provided in this application, this application has the following technical effects:

[0028] The present application provides a traction wheel wear fault diagnosis method based on digital twins, which initializes the population by using a method of generating pseudo-random numbers using a Halton sequence, optimizes the CNN-LSTM neural network by using a Pelican optimization algorithm based on a sine function to update the individual position, ensures the accuracy of CNN-LSTM neural network recognition, and performs rigid-flexible coupling dynamic simulation on the three-dimensional model of the traction elevator under the equivalent friction coefficient of each wire rope and traction wheel based on digital twin technology, obtains a data set, so that the simulation process can accurately reflect the actual running state of the traction wheel in real time, and uses the data set to train the optimized CNN-LSTM neural network to improve the accuracy of fault diagnosis model recognition, and can achieve high-precision traction wheel wear fault diagnosis effect. By inputting the car acceleration of the traction wheel to be diagnosed at the current moment, the wire rope tension of the car side wheel groove, and the wire rope tension of the counterweight side wheel groove into the fault diagnosis model, the wear fault diagnosis result of the traction wheel to be diagnosed can be obtained, which can achieve real-time and automated traction wheel wear fault diagnosis effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a schematic diagram of the contact between the wire rope and the traction wheel;

[0031] Figure 2 It is a cross-sectional view of a semicircular groove with a cutout;

[0032] Figure 3 This is a schematic diagram of the wheel groove angle before and after wear;

[0033] Figure 4 This is a schematic diagram of the analysis of the change in incision angle;

[0034] Figure 5 A flowchart of a traction sheave wear fault diagnosis method based on digital twin provided in one embodiment of the present application;

[0035] Figure 6 This is the structure diagram of the CNN-LSTM neural network;

[0036] Figure 7 A flowchart for obtaining a fault diagnosis model provided in one embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0038] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0039] The root cause of wire rope slip caused by traction sheave wear is as follows: Figure 1 shown.

[0040] When the elevator is loaded in the car and emergency brakes, the tension of the wire rope on both sides of the traction sheave always meets:

[0041]

[0042] Among them, T 1 Indicates the tension on the car side; T 2 It represents the tension on the counterweight side, e represents the natural base number, f represents the equivalent friction coefficient between the rope groove of the traction wheel and the wire rope, and α represents the wrap angle of the wire rope on the traction wheel.

[0043] like Figure 2 As shown in the figure, the traction capacity of the wheel groove is determined by the equivalent friction coefficient and the wrap angle of the wire rope on the wheel groove. The wear of the wheel groove will reduce the wrap angle, but the degree of reduction is not obvious and can be ignored. The equivalent friction coefficient of the semicircular groove of the traction wheel belt cutout is:

[0044]

[0045] Among them, γ is the unworn groove angle, c is the distance between the center of the circle and the upper edge of the groove when the groove is not worn. Considering the change in the depth of the traction wheel groove after wear, it can be seen from the relationship between the angles of the semicircular notched grooves that the overall center of the wire rope will drop after the wire rope and the traction wheel are worn, such as Figure 3 As shown, assuming that during the mutual running-in process between the wire rope and the traction wheel, if the wear depth of the wheel groove is ε, that is, the center of the fitting circle of the arc part of the wheel groove is lowered by ε, it is equivalent to the horizontal area of ​​the upper edge of the wheel groove being increased by ε.

[0046]

[0047] AC=ε·sin0.5γ (7)

[0048] Among them, γ′ represents the angle after wear, and d represents the traction sheave diameter.

[0049] Combining equations (3) to (7), we can get the relationship between the groove angle and the wear depth of the traction sheave groove:

[0050]

[0051] exist Figure 3 When the wheel groove wear reaches ε 0 When the outermost point D of the fitting circle of the wheel groove arc gradually wears to the straight segment of the wheel groove, the wheel groove angle γ′ decreases to zero. After that, the wheel groove continues to wear and the wheel groove angle γ′ no longer changes. At this time, the corresponding wear amount ε 0 for:

[0052]

[0053] exist Figure 4 In the figure, β is the initial cut angle of the notched semicircular wheel groove, β′ is the cut angle of the notched semicircular wheel groove after wear, and L is the cut width of the wheel groove. It can be found that the cut angle of the notched semicircular groove does not change with the wear of the traction sheave.

[0054] Therefore, the equivalent friction coefficient of the semicircular wheel groove with cutouts can be expressed as:

[0055]

[0056] μ represents the friction coefficient between the wire rope sheave grooves, and its value is 0.1 under loading conditions.

[0057] Based on this, in an exemplary embodiment, a traction wheel wear fault diagnosis method based on digital twin is provided. First, the rigid-flexible coupling dynamics simulation is performed on the three-dimensional model of the traction elevator under the equivalent friction coefficient of each wire rope and traction wheel. The digital twin method is used to enable the digital twin to accurately reflect the actual traction wheel operation status in real time. Secondly, the basic structure of the CNN-LSTM neural network is constructed, and the network hyperparameters are optimized through the IPOA algorithm to ensure its recognition accuracy. Finally, the real-time operation data is input into the neural network to complete the actual traction wheel operation status detection and fault diagnosis, so as to solve the fault diagnosis problem in the elevator operation process and reduce safety hazards. Figure 5 and Figure 7 As shown, the specific steps include:

[0058] Step 201: Under each equivalent friction coefficient between the wire rope and the traction wheel, a rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator to obtain sensor data corresponding to each wear amount during normal wear and abnormal wear under each equivalent friction coefficient to obtain a data set; the sensor data during normal wear includes the car acceleration during normal wear, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the sensor data during abnormal wear includes single-groove wear abnormal data and synchronous wear abnormal data; the single-groove wear abnormal data includes: the car acceleration during abnormal wear, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the synchronous wear abnormal data includes the car acceleration during abnormal wear, the wire rope tension of the target wheel groove on the car side, and the wire rope tension of the target wheel groove on the counterweight side.

[0059] Step 202: Use the IPOA algorithm to optimize the hyperparameters of the Long Short Term Memory Network (LSTM) in the CNN-LSTM neural network to obtain an optimized CNN-LSTM neural network. In the CNN-LSTM neural network, CNN is used to extract local feature information, LSTM is used to learn features, and Softmax is used for fault classification. The IPOA algorithm is a Pelican optimization algorithm that uses a Halton sequence to generate pseudo-random numbers to initialize the population, and uses an adaptive nonlinear weight factor based on a sine function to update the individual position; the population includes multiple individuals, and one individual includes all the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network. The IPOA algorithm is used to optimize the number of hidden layer neurons and the learning rate of the LSTM neural network; the optimization interval of the LSTM neural network hyperparameters is: the number of hidden layer neurons is [1,150], the initial learning rate is [0.001,0.2], the number of IPOA module populations is set to 30, and the maximum number of iterations is 20; the accuracy rate is selected as the fitness function; after the LSTM neural network is optimized by the IPOA algorithm, the optimal number of hidden layer neurons n is obtained. L and the learning rate L r .

[0060] Step 203: With the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as inputs and the wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained and tested according to the data set to obtain a fault diagnosis model; the wear fault diagnosis result is normal wear, single groove wear abnormality or synchronous wear abnormality.

[0061] Step 204: The car acceleration of the traction wheel to be diagnosed at the current moment, the wire rope tension of each wheel groove (5 wheel grooves) on the car side, and the wire rope tension of each wheel groove (5 wheel grooves) on the counterweight side (obtained by sensor measurement) are input into the fault diagnosis model to obtain the wear fault diagnosis result of the traction wheel to be diagnosed.

[0062] In another exemplary embodiment of the present application, under each equivalent friction coefficient between the wire rope and the traction wheel, a rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator, and sensor data corresponding to each wear amount during normal wear and abnormal wear under each equivalent friction coefficient is obtained to obtain a data set, which specifically includes:

[0063] Under any equivalent friction coefficient between the wire rope and the traction wheel, a rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration corresponding to each wear amount, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are obtained in the process of the groove angle wearing from the initial angle to the first preset angle at a first preset wear interval, and the sensor data corresponding to each wear amount in the normal wear process under the equivalent friction coefficient are obtained.

[0064] The rigid-flexible coupling dynamics simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration corresponding to each wear amount in the process of the groove angle wearing from the first preset angle to the second preset angle, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are obtained at the second preset wear amount interval, and the sensor data corresponding to each wear amount in the abnormal wear process under the equivalent friction coefficient are obtained; the first preset angle is greater than the second preset angle. The first preset angle is 25 degrees, and the second preset angle is 0 degrees.

[0065] The specific process of simulating the data set is as follows:

[0066] S11: Building a digital twin of the traction elevator: First, use SolidWorks software to build a 3D model of the traction elevator in proportion. The steel rope is modeled by SolidWorks combined curves. After the modeling is completed, it is converted into the Parasolid format x_t type and imported into RecurDyn for meshing and flexible body modeling. Components such as the traction wheel and the anti-ropes pulley are used as rigid bodies.

[0067] S12: Perform rigid-flexible coupling dynamics simulation in RecurDyn: Use RecurDyn software to continuously modify the equivalent friction coefficient between the wire rope and the traction wheel. γ greater than 25° is considered normal wear, γ less than 25° and greater than 0° is considered abnormal wear, and γ in Figure 2 The value drawn in the figure refers to the angle value of the groove, which is described in GB7588.12020. Taking the fully loaded down-going condition as the data collection for each simulation, taking γ as the initial groove angle of 40° as an example, when γ is worn to 25° (normal wear), the corresponding wear amount of the 10mm wire rope is 0.12mm; when γ is worn from 25° to 0° (abnormal wear), the corresponding wear amount of the 10mm wire rope is 0.88mm. The wear amount (0mm-0.12mm) is simulated every 0.002mm to obtain the tension of the five wheel groove wire ropes and the acceleration of the car, a total of 60 sets of normal wear data. The wear amount (0.126mm-0.88mm) is simulated every 0.01mm to obtain the tension of the five wheel groove wire ropes, the acceleration of the car, and 75 sets of single groove wear abnormality and synchronous wear abnormality data. And by constantly adjusting the simulation parameters, a high-fidelity virtual entity is constructed.

[0068] Data preprocessing involves processing missing values, outliers, data normalization, and standardization to obtain a data set.

[0069] This application uses RecurDyn multi-body dynamics software to build a rigid-flexible coupling model of the traction wheel and the wire rope, and continuously modifies the contact friction coefficient between the traction wheel and the wire rope to obtain the elevator wire rope tension and car acceleration during the fully loaded downward process as a data set, and uses tension and acceleration signals to reflect the degree of wear of the elevator. Then, the real-time collected sensor data is input into the fault diagnosis model to determine the current state of the traction wheel.

[0070] In another exemplary embodiment of the present application, the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove are used as inputs, and the wear fault diagnosis result is used as output. The optimized CNN-LSTM neural network is trained and tested according to the data set to obtain a fault diagnosis model, which specifically includes:

[0071] The data set is divided into a training set and a test set (divided into a training set and a test set in a ratio of 4:1).

[0072] The car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove are used as inputs, and the wear fault diagnosis results are used as outputs. The optimized CNN-LSTM neural network is trained according to the training set to obtain the trained CNN-LSTM neural network. During the training process, the training is stopped by judging whether the number of iterations i reaches the preset number of iterations T.

[0073] The trained CNN-LSTM neural network is tested using the test set to obtain the test results. The test set is input into the trained CNN-LSTM neural network, and the trained CNN-LSTM neural network will output the prediction results. The trained CNN-LSTM neural network is evaluated by the four indicators of accuracy, precision, recall, and F-measure to obtain the test results.

[0074] If the test result does not meet the requirements, the step of taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as input and the wear fault diagnosis result as output, and training the optimized CNN-LSTM neural network according to the training set to obtain the trained CNN-LSTM neural network is returned.

[0075] If the test results meet the requirements, the fault diagnosis model is obtained.

[0076] In another exemplary embodiment of the present application, Figure 6As shown in the figure, the CNN-LSTM neural network includes a convolutional neural network (CNN) and a long short-term memory neural network connected in sequence. The network parameter settings are as follows: the size of the convolution kernel in the CNN model is 64, the step size is 16, and the number of input channels is 11; the pooling kernel size of the pooling layer is 2, the step size is 2, and the padding method is 'same'. The number of nodes in the hidden layer of the first LSTM neural network is 128, the number of nodes in the hidden layer of the second LSTM neural network is 128, and the number of output nodes in the output layer using the softmax activation function is 3; the number of network training times is 200, and the initial learning rate is 0.005.

[0077] In another exemplary embodiment of the present application, the IPOA algorithm is used to optimize the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network, and the specific steps are:

[0078] The population is initialized according to the pseudo-random numbers generated by the Halton sequence to obtain multiple initial populations.

[0079] In the second stage of Pelican development, an adaptive nonlinear weight factor based on a sine function is introduced to update the position of individuals in the initial population to obtain the optimal individual. The optimal individual is determined as the optimal hyperparameter of the long short-term memory neural network, that is, the optimal number of hidden layer nodes n L and the learning rate L r , further improving the optimization performance of the Pelican algorithm.

[0080] In another exemplary embodiment of the present application, the use of randomly initialized populations in the Pelican Optimization Algorithm (POA) may result in insufficient diversity of the population, thereby reducing the search efficiency of the algorithm and slowing down the convergence of the algorithm. The present application uses the method of generating pseudo-random numbers using the Halton sequence to initialize the population, so that the algorithm has good ergodicity, so that it can be more evenly distributed during the solution process.

[0081] The specific steps of initializing the population according to the pseudo-random numbers generated by the Halton sequence are:

[0082] For the two-dimensional Halton sequence, by using two prime numbers p1 and p2 as the basis, the mathematical model of its segmentation process is as follows:

[0083]

[0084] θ(n)=b 0 p -1 +b 1 ·p -2…+b m ·p -m-1 (12)

[0085] H(n)=[θ 1 (n),θ 2 (n)] (13)

[0086] Where n is Halton's ordinal number. When calculating θ in formula (13), 1 (n), assign p in formula (11) to p1, and get n value n1. Then calculate θ(n1) according to formula (12), which is θ in formula (13). 1 (n), when calculating θ in formula (13) 2 (n), assign p in formula (11) to p2, and get n value n2. Then calculate θ(n2) according to formula (12), which is θ in formula (13). 2 (n), m is a real number with no practical meaning, p is a prime number greater than or equal to 2, b i ∈{0,1,…p-1} is a constant, θ(n) is the defined sequence function, H(n) is the obtained two-dimensional Halton sequence, i.e., the initial nth population, θ 1 (n),θ 2 (n) are the first and second individuals in the initial nth population. In practical applications, p1=2, p2=3.

[0087] Initialization of the pelican population:

[0088] q u,v = l v +H(n)·(d v -l v ) (14)

[0089] In the formula, q u,v represents the value of the vth hyperparameter in the uth individual in the initial population, l v d v are the lower and upper bounds of the vth hyperparameter respectively. This application optimizes the rand function through the Halton sequence.

[0090] During the optimization process, individuals can find the location of the optimal solution more quickly, thereby accelerating the convergence of the algorithm and improving the convergence accuracy of the algorithm.

[0091] In another exemplary embodiment of the present application, in the second stage of the existing Pelican optimization algorithm, according to the formula (15) Update the population, since there is a linearly decreasing term with the number of iterations This causes the algorithm to tend to the local optimum, which is inconsistent with the actual application. In order to solve this problem, the present application introduces an adaptive nonlinear weight factor based on a sine function in the sparrow search algorithm to improve the problem that the algorithm converges too quickly and falls into the local optimum in the middle and late stages of the iteration. The adaptive nonlinear weight factor based on the sine function is formulated as:

[0092] w=w max -(w max -w min )·sin((t / T) m ·(π / 2)) (16)

[0093] Replace w in formula (15) Update the population, where The value of the vth hyperparameter in the uth individual in the population after the second stage update, R = 0.2, w represents the adaptive nonlinear weight factor, w max Indicates the maximum inertia weight, w min represents the minimum inertia weight, m ​​is an adjustable parameter with no practical significance, which allows the w weight to change nonlinearly and improves the local search capability of the algorithm, t represents the number of cycles currently executed by the IPOA algorithm, T represents the maximum number of cycles of the IPOA algorithm, π represents the ratio of pi, and w max =0.9; w min =0.2;μ=2.

[0094] The simulation model of the present application realizes the real object state reflection of the wire rope and the traction wheel through the rigid-flexible coupling method. The model can effectively simulate the flexible deformation of the wire rope and the rigid characteristics of the traction wheel, providing more accurate simulation results. Through the simulation of the model, the reflection of the wire rope and the traction wheel under different working conditions can be analyzed, so as to effectively predict and diagnose potential faults in the elevator system. The accuracy and high fidelity of this model make the formulation of fault detection and maintenance strategies more scientific and reliable.

[0095] The present application can accurately diagnose the wear state of the traction sheave and provide a scientific basis for scientifically determining whether the traction sheave has failed. Moreover, the external acceleration and tension pressure sensors will not affect the normal operation of the elevator itself, and the measured data can deeply analyze the impact of wear on the failure of the traction sheave.

[0096] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0097] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0098] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above 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.

[0099] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A traction sheave wear fault diagnosis method based on digital twin, characterized in that: The traction sheave wear fault diagnosis method based on digital twin includes: Under each equivalent friction coefficient between the wire rope and the traction wheel, the rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator to obtain the sensor data corresponding to each wear amount during the normal wear process and the abnormal wear process under each equivalent friction coefficient, and obtain a data set; the sensor data during the normal wear process includes the car acceleration during the normal wear process, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the sensor data during the abnormal wear process includes the single-groove wear abnormal data and the synchronous wear abnormal data; the single-groove wear abnormal data includes: the car acceleration during the abnormal wear process, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side; the synchronous wear abnormal data includes the car acceleration during the abnormal wear process, the wire rope tension of the target wheel groove on the car side, and the wire rope tension of the target wheel groove on the counterweight side; The IPOA algorithm is used to optimize the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network to obtain the optimized CNN-LSTM neural network; the IPOA algorithm is a Pelican optimization algorithm that uses a Halton sequence to generate pseudo-random numbers to initialize the population and uses an adaptive nonlinear weight factor based on a sine function to update the individual position; the population includes multiple individuals, and one individual includes all the hyperparameters of the long short-term memory neural network in the CNN-LSTM neural network; With car acceleration, wire rope tension of car side wheel groove and counterweight side wheel groove as input and wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained and tested according to the data set to obtain the fault diagnosis model; the wear fault diagnosis result is normal wear, single groove wear abnormality or synchronous wear abnormality; The car acceleration of the traction sheave to be diagnosed at the current moment, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are input into the fault diagnosis model to obtain the wear fault diagnosis result of the traction sheave to be diagnosed.

2. The traction sheave wear fault diagnosis method based on digital twin according to claim 1 is characterized in that: Under each equivalent friction coefficient between the wire rope and the traction wheel, the rigid-flexible coupling dynamics simulation of the traction elevator three-dimensional model is carried out to obtain the sensor data corresponding to each wear amount in the normal wear process and the abnormal wear process under each equivalent friction coefficient, and obtain the data set, which specifically includes: Under any equivalent friction coefficient between the wire rope and the traction wheel, a rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration corresponding to each wear amount, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side are obtained in the process of the groove angle wearing from the initial angle to the first preset angle at a first preset wear amount interval, and the sensor data corresponding to each wear amount in the normal wear process under the equivalent friction coefficient are obtained; A rigid-flexible coupling dynamic simulation is performed on the three-dimensional model of the traction elevator. During the simulation, the car acceleration, the wire rope tension of each wheel groove on the car side, and the wire rope tension of each wheel groove on the counterweight side corresponding to each wear amount in the process of the groove angle wearing from the first preset angle to the second preset angle are obtained at a second preset wear interval to obtain sensor data corresponding to each wear amount in the abnormal wear process under the equivalent friction coefficient; the first preset angle is greater than the second preset angle.

3. The traction sheave wear fault diagnosis method based on digital twin according to claim 2 is characterized in that: The first preset angle is 25 degrees, and the second preset angle is 0 degrees.

4. The traction sheave wear fault diagnosis method based on digital twin according to claim 1 is characterized in that: Taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as input and the wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained and tested according to the data set to obtain the fault diagnosis model, which specifically includes: Dividing the data set into a training set and a test set; Taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as inputs and the wear fault diagnosis result as output, the optimized CNN-LSTM neural network is trained according to the training set to obtain the trained CNN-LSTM neural network; The trained CNN-LSTM neural network is tested using the test set to obtain the test results; If the test result does not meet the requirements, the step of taking the car acceleration, the wire rope tension of the car side wheel groove and the wire rope tension of the counterweight side wheel groove as inputs and the wear fault diagnosis result as output, and training the optimized CNN-LSTM neural network according to the training set to obtain the trained CNN-LSTM neural network is returned; If the test results meet the requirements, the fault diagnosis model is obtained.

5. The traction sheave wear fault diagnosis method based on digital twin according to claim 1, characterized in that: The CNN-LSTM neural network consists of a convolutional neural network and a long short-term memory neural network connected in sequence.

6. The traction sheave wear fault diagnosis method based on digital twin according to claim 1, characterized in that: The adaptive nonlinear weight factor based on the sine function is: w=w max -(w max -w min )·sin((t / T) m ·(π / 2)), where w represents the adaptive nonlinear weight factor, w max Indicates the maximum inertia weight, w min It represents the minimum value of inertia weight, m ​​is a real number and has no practical meaning, t represents the number of loops currently executed by the IPOA algorithm, T represents the maximum number of loops of the IPOA algorithm, and π represents the ratio of pi.

Images