Fault prediction method, device and product for large shaft of recreation facility

Abstract

The invention discloses a recreation facility large-scale shaft fault prediction method, device and product, and relates to the field of fault prediction, and the method comprises the steps: obtaining the vibration data of a recreation facility large-scale shaft; the amusement facility large-scale shaft vibration data comprises a main shaft rotation speed per hour, a vibration amplitude, a main shaft temperature, a vibration frequency, an acceleration, a torque, a load weight, power consumption and a data acquisition time point; performing feature extraction on the large-scale shaft vibration data of the recreation facility to obtain extracted features; carrying out dimension reduction processing on the extracted features to obtain features after dimension reduction; performing classification prediction on the dimensionality-reduced features to obtain classification categories; the classification categories comprise bearing damage, load imbalance, looseness, rolling body damage and other faults. The device state can be monitored in real time, future possible faults can be predicted, and the accuracy and the real-time performance of fault prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of fault prediction, and in particular to a method, equipment and product for predicting faults of large shafts of amusement facilities. Background Art

[0002] In the IoT environment, data collection is performed by integrating multiple sensors and using advanced data analysis technology to predict and diagnose potential equipment failures. With the widespread use of large-scale entertainment facilities, their safe operation is particularly important. As a facility that is highly dependent on mechanical and structural integrity, the operating status of its large shafts is directly related to the safety and stability of the entire facility.

[0003] In traditional maintenance strategies, equipment maintenance and fault diagnosis often rely on regular inspections or intervention only after signs of failure appear. This approach is not only time-consuming and labor-intensive, but also difficult to prevent before it happens. Therefore, it is very necessary to develop a technology that can monitor the status of equipment in real time and predict possible future failures. With the development and application of Internet of Things technology, it is now possible to optimize the maintenance and fault prevention strategy of Ferris wheels through real-time data collection and analysis. The Internet of Things platform can integrate a variety of sensors to monitor the operating status of large axes of amusement facilities in real time, such as vibration, temperature, and load. These data are transmitted to the central processing unit in real time through wireless networks for storage and analysis. However, faced with massive amounts of data and complex data features, how to extract useful information from them and accurately predict potential failures is a technical challenge. In addition, existing data analysis methods often have problems of low efficiency and low accuracy when processing high-dimensional data. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device and product for predicting large shaft faults of amusement facilities, which can improve the accuracy and real-time performance of fault prediction.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] In a first aspect, the present invention provides a method for predicting a large shaft fault of an amusement facility, the method comprising:

[0007] Obtain vibration data of a large shaft of an amusement facility; the vibration data of a large shaft of an amusement facility includes: main shaft speed per hour, vibration amplitude, main shaft temperature, vibration frequency, acceleration, torque, load weight, power consumption and data collection time point.

[0008] Feature extraction is performed on the vibration data of the large shaft of the amusement facility to obtain extracted features.

[0009] Performing dimensionality reduction processing on the extracted features to obtain features after dimensionality reduction.

[0010] Classification prediction is performed on the features after dimension reduction to obtain classification categories; the classification categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults.

[0011] In a second aspect, the present invention provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for predicting large shaft faults of amusement facilities.

[0012] In a third aspect, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned method for predicting large shaft failures of amusement facilities.

[0013] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0014] The present invention provides a method, device and product for predicting the fault of a large shaft of an amusement facility. By acquiring the vibration data of a large shaft of an amusement facility, the feature extraction is performed on the vibration data of the large shaft of the amusement facility to obtain the extracted features, thereby obtaining the representative features in the vibration data of the large shaft of the amusement facility; by performing dimensionality reduction processing on the extracted features to obtain the reduced-dimensional features, the complexity of data processing and storage overhead can be reduced; and then the reduced-dimensional features are classified and predicted to obtain classification categories. Compared with the prior art, the present invention can monitor the equipment status in real time and predict possible future faults, and improves the accuracy and real-time performance of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A schematic flow chart of a method for predicting large shaft faults of an amusement facility provided in one embodiment of the present invention.

[0017] Figure 2 A schematic diagram of the training process of a neural network parameter optimization algorithm based on the fusion phenomenon principle provided by one embodiment of the present invention.

[0018] Figure 3 A schematic diagram of the structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] 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.

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] The Chinese invention patent with application number CN201911126775.2 proposes an early diagnosis method for bearing faults using adaptive frequency-shifted variable-scale vibration resonance technology. In this method, the original vibration signal of the bearing is first extracted, and then the bearing vibration signal is processed by a signal processing function to generate a superposition signal. Subsequently, a quantum genetic algorithm is used to adjust the parameters of the bistable state function for the purpose of improving the signal-to-noise ratio to optimize the superposition signal, thereby achieving a better output waveform. Beneficial effect: This method can use the collected signal to quickly optimize the steady-state function parameters through a quantum genetic algorithm, improve the signal-to-noise ratio of the output signal, and thus use noise to enhance the bearing fault characteristic signal, effectively detect bearing faults, and the detection effect is better than the traditional maximum correlation spectrum kurtosis deconvolution method.

[0022] The Chinese invention patent with application number CN202110175492.8 proposes a method, device, medium and computer equipment for rolling bearing fault diagnosis, including: intercepting the vibration signal of the rolling bearing to obtain an initial signal; creating an initial Hankel matrix based on the initial signal, and reconstructing the initial Hankel matrix using a singular value decomposition algorithm to obtain a reconstructed signal; demodulating the reconstructed signal using a 1.5-dimensional symmetric difference analytical energy operator demodulation algorithm to obtain a demodulated signal; determining the 1.5-dimensional energy spectrum of the fault feature signal based on the demodulated signal; diagnosing bearing faults based on the 1.5-dimensional energy spectrum; in this way, the original vibration signal of the rolling bearing is reconstructed using an adaptive singular value decomposition algorithm to remove background noise in the vibration signal and obtain fault features; then processing the fault features using a 1.5-dimensional symmetric difference analytical energy operator demodulation algorithm to suppress residual noise, improve fault features, and ensure the accuracy of fault diagnosis results.

[0023] The Chinese invention patent with application number CN201910159895.6 proposes a rolling bearing fault diagnosis method under variable working conditions based on deep features and transfer learning, which relates to the field of fault diagnosis technology. It aims to solve the problem of low accuracy of multi-state recognition of different fault locations and different performance degradation degrees of rolling bearings under variable working conditions. First, SDAE is used to extract the frequency domain amplitude of the rolling bearing vibration signal to obtain the deep features of the vibration signal, and form a source domain feature sample set and a target domain feature sample set. Then, JGSA is used to perform domain adaptation processing on the source domain feature samples and the target domain feature samples, so as to simultaneously reduce the distribution offset and subspace transformation differences of the feature samples between domains, thereby reducing the domain offset between feature samples of different types. Finally, the multi-state classification of rolling bearings under variable working conditions is completed by the K nearest neighbor algorithm. Compared with other methods, this invention shows better feature extraction ability under variable working conditions of rolling bearings, has the best sample feature visualization effect, and has high accuracy of rolling bearing fault diagnosis under variable working conditions.

[0024] However, the following problems in the above technical solution still need to be further solved:

[0025] 1. Traditional data collection may be limited by incomplete sensor coverage, low data collection frequency, or limited data types, resulting in insufficient data volume or low data quality for training models.

[0026] 2. Traditional neural network optimization methods (such as gradient descent) may encounter gradient vanishing or exploding problems, especially in complex network structures, which will affect the stability and performance of the model.

[0027] 3. Existing feature extraction and dimensionality reduction technologies may not be able to effectively process high-dimensional data, resulting in the loss of important information and affecting the accuracy of fault identification.

[0028] 4. Traditional classification algorithms may be inefficient in parameter optimization and searching solution space, especially in large-scale parameter spaces, where it is difficult to find the optimal solution.

[0029] 5. Existing systems may lack sufficient real-time data processing capabilities and adaptability to cope with rapidly changing operating conditions and emerging fault types.

[0030] In order to solve these problems, the present invention proposes a classification algorithm based on advanced machine learning technology, including generative adversarial networks, neural network optimization based on the principle of fusion phenomenon, autoencoder dimensionality reduction, and heterogeneous bee colony optimization, so as to improve the accuracy and real-time performance of fault prediction.

[0031] In an exemplary embodiment, Figure 1 As shown, a method for predicting large shaft faults of amusement facilities is provided, and the method includes the following steps S1 to S4. Among them:

[0032] S1: Acquire the vibration data of the large shaft of the amusement facility; the vibration data of the large shaft of the amusement facility includes: the main shaft speed per hour, the vibration amplitude, the main shaft temperature, the vibration frequency, the acceleration, the torque, the load weight, the power consumption and the data collection time point.

[0033] S2: extracting features from the large shaft vibration data of the amusement facility to obtain extracted features;

[0034] S3: performing dimensionality reduction processing on the extracted features to obtain features after dimensionality reduction;

[0035] S4: Classify and predict the features after dimension reduction to obtain classification categories; the classification categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults.

[0036] Implementing the above steps S1 to S4 can improve the accuracy and real-time performance of fault prediction.

[0037] In another exemplary embodiment of the present invention, in step S2, specifically including:

[0038] The vibration data of the large shaft of the amusement facility is input into a feature extraction model to obtain extracted features; the feature extraction model is a neural network parameter optimization algorithm based on the principle of fusion phenomenon.

[0039] like Figure 2 As shown, the method for determining the feature extraction model is:

[0040] S21: Acquire historical vibration data of large shafts of amusement facilities; the historical vibration data of large shafts of amusement facilities include: main shaft speed per hour, vibration amplitude, main shaft temperature, vibration frequency, acceleration, torque, load weight, power consumption and data collection time point.

[0041] The data of the present invention is collected on an Internet of Things platform, which integrates multiple sensors to monitor and record the vibration parameters of large shafts of amusement facilities. The data is sent to a central processing unit in real time via wireless transmission and stored in a specific storage format, specifically the CSV format.

[0042] As an optional implementation, the attributes of the data include:

[0043] a rph : Spindle speed per hour.

[0044] a vib : Vibration amplitude.

[0045] a temp : Spindle temperature.

[0046] a freq : Vibration frequency.

[0047] a accl : acceleration.

[0048] a torq : Torque.

[0049] a load : Load weight.

[0050] a power : Power consumption.

[0051] a time : Data collection time point.

[0052] In this embodiment, one of the data records is as follows:

[0053] a rph =2.

[0054] a vib =0.5mm / s.

[0055] a temp =60℃.

[0056] a freq =150Hz.

[0057] a accl =9.8m / s 2 .

[0058] a torq =300Nm.

[0059] a load =2000t.

[0060] a power =44kW.

[0061] a time ='2024-04-24T08:30:00Z'.

[0062] It should be emphasized that this embodiment is only used to illustrate a data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds.

[0063] S22: annotate the vibration data of the historical amusement facility large shaft to obtain annotated data; the annotated categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults. The annotating method of this embodiment is manual annotating.

[0064] S23: Based on the labeled data, a data expansion model is used to expand the vibration data of the large axis of the historical amusement facility to obtain expanded data; the data expansion model is a quantum state feedback generative adversarial network algorithm.

[0065] The expanded data is input into the feature extraction model to train the feature extraction model. The present invention uses a 3-layer fully connected neural network for feature extraction. In the formula, some schemes use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention adopts a neural network parameter optimization algorithm based on the principle of fusion phenomenon. Inspired by the fusion phenomenon in nature, lighter atomic nuclei merge under high temperature and high pressure to form more stable heavy atomic nuclei, while releasing huge energy. The neural network parameter optimization algorithm based on the principle of fusion phenomenon optimizes the weights and biases of the network by simulating the fusion process. The weights and biases are regarded as atomic nuclei, and their merging and transformation processes are intended to achieve the optimal state of the loss function. Unlike the traditional optimization algorithm based on gradient descent, the neural network parameter optimization algorithm based on the principle of fusion phenomenon does not rely on gradient information, thereby avoiding the problem of gradient vanishing or exploding.

[0066] S24: Initializing the parameters of the neural network according to the expanded data to obtain initialized neural network parameters.

[0067] Define the structure of the neural network. The three-layer fully connected neural network used includes an input layer and two hidden layers, and a preset output layer for category supervision training, which are defined as follows:

[0068] Input layer, assuming the dimension of the input vector is d, then the input layer has d neurons, corresponding to the d elements in the vector.

[0069] Hidden layers, both hidden layers have 128 neurons and use ReLU as the activation function.

[0070] Output layer,The output layer uses the softmax function as the activation function.

[0071] Furthermore, the neural network parameters are initialized and assigned a random energy state. Specifically, the weights w and biases b of the neural network are assigned initial values, and the energy state E(w,b) is defined to describe the potential optimization ability of the parameters, which can be expressed as:

[0072]

[0073] In the formula, x i represents the input data (expanded data), y irepresents the corresponding target output (the label corresponding to the expanded data), N represents the number of samples in the data set, K represents the number of feature scales, and α k (t) represents the adaptive weight of the k-th scale, which depends on the training time t and the current performance state of the network, ensuring that the focus of the loss function is adjusted at different stages of training.

[0074] Furthermore, α k (t) is dynamically updated at each iteration, which allows the model to focus on different feature levels at different training stages, thereby more effectively learning the multi-level structure of complex data. Specifically, the weight α of each scale k The update of (t) depends on the effect of the scale under the current training state. The update method of the adaptive weight can be expressed as:

[0075]

[0076] In the formula, E k (t) represents the loss value of the kth scale at time t; β k It represents the importance adjustment coefficient of the kth scale, which is preset artificially according to the characteristics of the scale.

[0077] It should be noted that the output layer is only used for energy state monitoring during training, and the output of the feature extraction model is the output of the second hidden layer of the neural network, that is, the feature vector after feature extraction.

[0078] S25: Perform fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set.

[0079] Perform a nuclear synthesis cycle operation, and perform a fusion candidate check on each pair of parameters (weight and bias). If the synthesis of the two parameters can reduce the overall system energy, fusion is allowed, otherwise it remains independent. Specifically, for each pair of parameters w j and b k An assessment is made to determine whether the total energy of the system can be reduced through fusion, which can be expressed as:

[0080] ΔE=E(w j ',b k ')-E(w j ,b k );

[0081] And, the fusion behavior of the parameters can be expressed as:

[0082] w j '=w j +δ tr ;

[0083] b k '=bk +∈ tr ;

[0084] In the formula, w j 'and b' k are the updated neural network parameters respectively. If ΔE<Θ(t), the new parameter w is accepted. j 'and b' k , otherwise keep the original parameters unchanged, Θ(t) is the energy threshold of the current iteration.

[0085] Furthermore, the energy threshold Θ(t) is updated at each iteration. The energy changes during the training process are monitored in real time, and the energy threshold is adjusted dynamically to control the sensitivity and frequency of parameter updates. Specifically, the update rule of the energy threshold is defined as:

[0086]

[0087] Where Θ(t) and Θ(t+1) are the energy thresholds for the current and next time steps respectively; ν eo is the learning rate of threshold adjustment, which controls the speed of adjustment; E(t) is the average energy of the current time step; E target is the target energy, which is preset according to the task requirements; σ E is the standard deviation of energy variation; tanh is the hyperbolic tangent function. Preferably, ν eo Set to 0.01, E target Set to 0.1.

[0088] Furthermore, the weight adjustment amount δ tr and bias adjustment ∈ tr The calculation method can be expressed as:

[0089]

[0090]

[0091] In the formula, is the gradient of the energy value with respect to the weight parameter, is the gradient of the energy value with respect to the bias parameter, σ to and η to is the sensitivity and scaling factor of the fusion adjustment. Preferably, σ to and η to Set them to 0.1 and 0.3 respectively.

[0092] S26: Determine the energy state of the synthesized parameter set.

[0093] The energy of the synthesized new parameter set is evaluated, the parameter state is updated, and the parameter group in the high energy state is pushed to the low energy state to stabilize the desired state. Specifically, the synthesized parameter set is re-evaluated through the energy function E to ensure the energy is minimized, which can be expressed as:

[0094]

[0095] In the formula, p(y i |x i ; w, b) means that given the input x i With parameters w, b, output y i probability.

[0096] Furthermore, logp(y i |x i ; The calculation method of w, b) can be expressed as:

[0097]

[0098] In the formula, σ std Indicates the standard deviation of the output data.

[0099] S27: Adjust the nuclear reaction temperature according to the energy change.

[0100] According to the current network performance and stability (energy change), the nuclear reaction temperature is adjusted to simulate the temperature control in physical fusion and adjust the fusion temperature T, which can be expressed as:

[0101] T new =T old ·exp(-α te ΔE);

[0102] In the formula, α te is the temperature adjustment factor, T new and T old are the temperatures before and after adjustment, and ΔE is the energy change.

[0103] Furthermore, α te Based on dynamic adjustment of energy changes, its calculation method can be expressed as:

[0104]

[0105] In the formula, κ te is a preset adjustment coefficient, |ΔE| is the absolute energy change, E avg is the average value of all current parameter energies.

[0106] S28: Determine whether the energy state of the synthesized parameter set reaches a dynamic equilibrium state, and obtain a first determination result.

[0107] S29: If the first judgment result is no, return to the step of "performing a fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set".

[0108] S210: If the first judgment result is yes, update the parameter configuration based on energy optimization to obtain updated neural network parameters.

[0109] Detect whether the energy of the network parameters reaches a dynamic equilibrium state. If equilibrium is reached, continue the optimization process; if not, perform parameter re-fusion processing. In one embodiment, the judgment condition can be expressed as:

[0110] Var(E)<∈ mr ;

[0111] Where Var(E) is the variance function of the energy E(w,b), preferably, ∈ mr Set to 10 -5 .

[0112] S211: Determine whether the energy state of the updated neural network parameters converges to obtain a second determination result.

[0113] S212: If the second judgment result is no, return to the step of "performing a fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set.

[0114] S213: If the second judgment result is yes, output the final optimized neural network parameters to obtain a feature extraction model.

[0115] According to the energy optimization principle, whether to update the current parameter configuration can be expressed as:

[0116]

[0117] b new =b k '+β rate Δb;

[0118] In the formula, β rate To update the learning rate, is the fusion adjustment factor. Preferably, the learning rate β is set to 0.001.

[0119] Furthermore, the weight and bias parameters are updated using momentum terms, and the calculation method can be expressed as:

[0120]

[0121]

[0122] In the formula, μ ea is the momentum coefficient, Δw prev and Δb prev are the weight and bias updates of the previous iteration respectively.

[0123] Furthermore, the fusion adjustment factor According to the historical changes of weights and the error rate of the current layer, it is dynamically set and can be expressed as:

[0124]

[0125] In the formula, is the difference between the current weight and the weight of the previous iteration; is the standard deviation of the weight change of the current layer; E l is the error rate of the current layer; ρ yt and λ yt is the adjustment coefficient, which controls the influence intensity of weight change rate and error rate respectively; tanh is the hyperbolic tangent function. Preferably, ρ yt and λ yt Set them to 0.3 and 0.5 respectively.

[0126] The convergence of the algorithm is determined by the change in parameter energy after multiple iterations. If converged, the final optimized parameters are output; if not, return to step S25 to continue iteration. The convergence judgment method can be expressed as:

[0127]

[0128] In the formula, E new is the energy state after parameter update, E old is the energy state before parameter update, γ ew Preferably, the convergence threshold γ ew Set to 10 -6 .

[0129] In step S23, it can be understood that in the task of the present invention, the acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability, while affecting the accuracy of the model. The present invention adopts a quantum state feedback generative adversarial network algorithm for sample generation, thereby achieving data expansion. The quantum state feedback generative adversarial network algorithm consists of two parts: a generator (G) and a discriminator (D). The purpose of the generator is to create realistic data samples, and the discriminator attempts to distinguish between generated samples and real samples. In traditional generative adversarial network models, the training of generators and discriminators usually depends on a dynamic balance of mutual competition. The present invention allows the generator to receive a special quantum state feedback during the generation process by adopting the principle of quantum teleportation, and the quantum state feedback carries additional information about the internal state of the discriminator. In addition, when evaluating the data, the discriminator will also adjust its loss function according to the change of the quantum state to enhance the adaptability and generalization ability of the model to new data features.

[0130] The method for determining the data expansion model is:

[0131] S2301: Initialize the quantum state parameters, the initial quantum state, the network parameters of the generator, and the network parameters of the discriminator according to the labeled data to obtain the initialized quantum state parameters, the initial quantum state, the network parameters of the generator, and the network parameters of the discriminator.

[0132] Initialize the quantum state parameters and the initial quantum state, and initialize the network parameters of the generator and the discriminator at the same time. Preferably, the quantum state parameters are randomly initialized by Gaussian distribution, and the weights (network parameters) of the generator and the discriminator are set by the Xavier initialization method.

[0133] Specifically, the method of initializing quantum state parameters can be expressed as:

[0134]

[0135] In the formula, ψ binit represents the initial quantum state vector, Represents a Gaussian distribution with mean 0 and a covariance matrix that is the identity matrix.

[0136] Moreover, the parameters of the generative adversarial network (network parameters of the generator and the discriminator) can be initialized as follows:

[0137]

[0138]

[0139] In the formula, θ bG and θ bD Represent the parameters of the generator and the discriminator respectively; Indicates uniform distribution in the interval [a, b]; n bin and n bout Represents the number of nodes in the network input and output layers respectively.

[0140] S2302: Generate quantum state code according to the initialized quantum state parameters and the initial quantum state;

[0141] Generate a set of quantum states, each corresponding to a potential input space, and assign the quantum state to the generator as additional input information. Furthermore, the generator receives the actual quantum information converted from the quantum state parameters as part of the input. The quantum state encoding method can be expressed as:

[0142] φ benc =tanh(W bφ ψ binit +b bφ );

[0143] In the formula, φ benc is the encoded quantum state; W bφ and b bφ are the weight matrix and bias vector of quantum state encoding respectively; tanh is the activation function.

[0144] In one embodiment, to increase the nonlinearity and adaptability of the model, the weight matrix W bφ The calculation method can be expressed as:

[0145]

[0146] Where V bφ and U bφ is the orthogonal matrix obtained by singular value decomposition, s bφ is a scaling vector that is randomly initialized and optimized by gradient descent, and diag() is used to convert the vector into a diagonal matrix.

[0147] S2303: Generate a virtual data sample according to the quantum state encoding and input random noise;

[0148] The generator generates virtual data samples based on the input random noise and the corresponding quantum state information. The information of the quantum state is used to guide the generator to output more diverse and realistic data samples. The way the generator generates data samples can be expressed as:

[0149]

[0150] In the formula, x bgen represents the generated data sample; is a generator function; is random noise.

[0151] S2304: According to the virtual data sample, a discriminator is used to provide a discriminant result;

[0152] The discriminator evaluates the data samples from the generator and the real data samples, giving the probability of being a real sample. The way the discriminator discriminates the data can be expressed as:

[0153]

[0154] In the formula, represents the discriminator function; Sig() is the Sigmoid activation function; W bD and b bD are the weight matrix and bias vector of the discriminator.

[0155] S2305: Adjust the quantum state parameters according to the discrimination result, the real sample label, and the loss function of the discriminator;

[0156] According to the discrimination results, the real sample labels, and the loss function of the discriminator, the quantum state feedback mechanism is optimized. The quantum state adjustment method can be expressed as:

[0157]

[0158] In the formula, Δψ represents the adjustment amount of the quantum state; λ badj Indicates the adjustment strength; Represents ψ binit The gradient of represents the loss function of the discriminator; y breal represents the real sample label; α bψ Represents the learning rate.

[0159] In one embodiment, considering the integration of quantum information, the discriminator loss function The calculation method can be expressed as:

[0160]

[0161] In the formula, y breal is the label of the real sample, β breg is the regularization coefficient, is the L2 norm of the quantum state encoding.

[0162] S2306: Update the network parameters of the generator and the discriminator using the gradient descent method;

[0163] Use the gradient descent algorithm to update the parameters of the generator and discriminator. The way to update the parameters can be expressed as:

[0164]

[0165]

[0166] In the formula, γ bG and γ bD is the learning rate; and Represent the gradients of the generator and discriminator parameters respectively; and are the loss functions of the generator and the discriminator respectively; y bfake is to generate sample labels. Preferably, γ bG and γ bD Set to 0.001 and 0.0005 respectively.

[0167] Furthermore, the gradient of the generator with respect to the loss function The calculation method can be expressed as:

[0168]

[0169] In the formula, λ bdec is the weight decay coefficient. Preferably, λ bdec Set to 0.1.

[0170] S2307: Determine whether the condition for stopping iteration is met, and obtain a third determination result;

[0171] S2308: If the third judgment result is no, return to the step of “generating a virtual data sample according to the quantum state encoding and the input random noise”;

[0172] S2309: If the third judgment result is yes, the final quantum state parameters, network parameters of the generator and network parameters of the discriminator of the quantum state feedback generative adversarial network algorithm are determined to obtain a data expansion model.

[0173] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0174] In another exemplary embodiment of the present invention, the data after feature extraction is input into the feature dimension reduction model to train the feature dimension reduction model. The present invention adopts an autoencoder algorithm for feature dimension reduction, and the autoencoder algorithm consists of two parts: an encoder and a decoder. The encoder is responsible for converting the input high-dimensional feature data into a lower-dimensional representation, and the decoder attempts to restore the original data from these low-dimensional representations. Based on the traditional autoencoder, the present invention adopts the Markov chain Monte Carlo method to optimize the training process of the encoder and decoder, thereby improving the efficiency and accuracy of feature dimension reduction.

[0175] In step S3, it specifically includes:

[0176] S31: Input the extracted features into a feature dimensionality reduction model to obtain reduced features; the feature dimensionality reduction model is an autoencoder algorithm.

[0177] The method for determining the feature dimension reduction model is:

[0178] S3101: Initialize the parameters of the autoencoder algorithm to obtain initialized encoder parameters and decoder parameters.

[0179] Initialization: Initialize the parameters of the autoencoder model, which include the weights and biases of the encoder and decoder. Preferably, small random numbers are used for initialization to avoid symmetry destruction and gradient vanishing problems. Specifically, the parameter initialization method can be expressed as:

[0180] θ pE ,θ pD ←Xavier(size pE ,size pD );

[0181] In the formula, θ pE and θ pD Respectively represent the parameters of the encoder and decoder; Xavier() represents the Xavier parameter initialization method, size pE and size pD denote the network sizes of the encoder and decoder respectively.

[0182] S3102: Generate a series of samples using the Markov Chain Monte Carlo method; the samples include a number of historically extracted features; the historically extracted features are features obtained by inputting the expanded data into a feature extraction model.

[0183] Before each round of training, a series of samples are generated using the Markov chain Monte Carlo method. These samples are used to train the autoencoder of this round. Moreover, these samples are based not only on the original data, but also on the data generated by the model in the previous round of training. The method of generating Markov chain samples can be expressed as:

[0184]

[0185]

[0186] In the formula, represents the sample of the t+1th iteration; represents the sample of the tth iteration; σ pM Represents the iteration step parameter; ∈ (t)Represents a normal distribution The noise extracted from .

[0187] S3103: Input the sample into an autoencoder algorithm to obtain an output of the autoencoder algorithm, where the output of the autoencoder algorithm is reconstructed data corresponding to the sample.

[0188] The data is compressed into low-dimensional features by the encoder, and then the decoder attempts to restore the original data. The encoding and decoding process can be expressed as:

[0189]

[0190]

[0191] In the formula, z p It is the encoded low-dimensional feature representation; and are the mapping functions of the encoder and decoder respectively; x p is the input data; is the reconstructed data.

[0192] Furthermore, the calculation method of the encoder mapping function can be expressed as:

[0193]

[0194] Where W pE and b pE Represent the weight matrix and bias vector of the encoder respectively; Sig() represents the Sigmoid activation function.

[0195] And, the calculation method of the decoder mapping function can be expressed as:

[0196]

[0197] Where W pD and b pD are the weight matrix and bias vector of the decoder respectively.

[0198] S3104: Determine a loss value according to the sample, the reconstructed data corresponding to the sample and the loss function.

[0199] Furthermore, the loss function measures the difference between the input data and the output data by calculating the reconstruction error. The calculation method of the loss function can be expressed as:

[0200]

[0201] In the formula, is the total loss; pR is the regularization strength, represents the L2 norm. Preferably, λ pR Set to 0.4.

[0202] The gradient of the parameters is calculated through the back propagation algorithm according to the loss function. The way of updating the parameters during the back propagation process can be expressed as:

[0203]

[0204]

[0205] In the formula, α p is the learning rate; and is the loss gradient of the encoder and decoder parameters. Preferably, α p Set to 0.001.

[0206] Furthermore, the gradient and The calculation method can be expressed as:

[0207]

[0208]

[0209] In the formula, is the loss for the encoded feature z p The partial derivative of is the partial derivative of the encoded feature with respect to the encoder parameters; is the loss to reconstruct the data The partial derivative of is the partial derivative of the reconstructed data with respect to the decoder parameters.

[0210] S3105: According to the loss value, the encoder parameters and the decoder parameters of the autoencoder algorithm are optimized to obtain a feature dimensionality reduction model.

[0211] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0212] In another exemplary embodiment of the present invention, the reduced-dimensional data is input into a classifier to train the classifier model. The present invention adopts an elastic net classification algorithm based on heterogeneous swarm optimization as a classifier. Quantum coding technology is used in the heterogeneous swarm optimization algorithm. The state of each solution is represented by a quantum bit, and a search can be performed in a wider solution space to find better model parameters.

[0213] In step S4, it specifically includes:

[0214] S41: Inputting the dimension-reduced features into a classifier model to obtain classification categories; the classifier model is an elastic net classification algorithm based on heterogeneous bee colony optimization.

[0215] The method for determining the classifier model is:

[0216] S4101: Initialize the quantum bit state, position and speed of each swarm member according to the features after historical dimensionality reduction, and obtain the initialized quantum bit state, position and speed of each swarm member; the features after historical dimensionality reduction are the features obtained by inputting the features after historical extraction into the feature dimensionality reduction model.

[0217] Initialize the position and velocity of the swarm, and each swarm member represents a possible classifier parameter setting. Specifically, use qubits to initialize each solution so that the initial state of each solution contains both determinism and probability to enhance the global search capability. The way to initialize qubits and parameters can be expressed as:

[0218]

[0219] In the formula, represents the initial state of the kth qubit.

[0220] Furthermore, the method of initializing the swarm position and velocity can be expressed as:

[0221]

[0222]

[0223] In the formula, and are the initial position and velocity of the i-th swarm member; rand(bound q ) indicates that within a given boundary bound q Generate random numbers to initialize position and velocity.

[0224] S4102: Determine the performance index of each swarm member according to the position of each swarm member after initialization.

[0225] The performance of each swarm member is evaluated, that is, the model is trained using the reduced feature vector and the corresponding label, and the accuracy of the model is calculated. The performance evaluation method can be expressed as:

[0226]

[0227] Where P qirepresents the performance index of the ith swarm member, calculated by the evaluate() function, which is based on the classifier using x qi The classification accuracy on the validation set as a parameter.

[0228] S4103: Determine the location of the swarm member with the best performance index.

[0229] Sorting the swarm members according to their performance, selecting the best performing individuals to enter the next round of iteration, the method of selecting the optimal solution can be expressed as:

[0230]

[0231] In the formula, x qbest is the position of the best performing swarm member in the current iteration.

[0232] S4104: Based on the position of the swarm member with the best performance index, a heterogeneous swarm optimization strategy is adopted to update the quantum bit state, position and speed of each swarm member.

[0233] A heterogeneous swarm optimization strategy is adopted to update the position and speed of swarm members by imitating the behavior of bees in nature. The heterogeneous swarm optimization strategy includes the behavior of followers and scouts, as well as the probability adjustment of qubits. Specifically, the way to update qubits and positions during the swarm update process can be expressed as:

[0234]

[0235]

[0236] In the formula, β q is a mixing parameter adjusted based on performance feedback, β q is based on the current individual performance P qi Relative to the average performance of the colony P qavg The difference of the adjusted q is a parameter that controls the sensitivity of feedback, and its value range is between 0.1 and 1.0. Preferably, γ q Set to 0.45, P qavg is the average performance of the current swarm.

[0237] According to the currently searched optimal solution, the state of the qubit is updated to reflect the latest search information and improve the algorithm's search efficiency in unexplored areas. The speed and position update method can be expressed as:

[0238]

[0239]

[0240] In the formula, ωq is the inertia weight, φ q1 ,φ q2 is the learning factor, r q1 , r q2 is a random number in the range [0, 1], x qgbest is the global optimal position. Preferably, ω q Set to 0.3, φ q1 ,φ q2 Set to 0.5 and 0.2 respectively.

[0241] S4105: Determine whether the condition for stopping iteration is met, and obtain a fourth determination result.

[0242] S4106: If the fourth judgment result is no, return to the step of "determining the performance index of each swarm member according to the initialized position of each swarm member".

[0243] S4107: If the fourth judgment result is yes, the position of the swarm member with the best performance index is determined as the optimal model parameter to obtain a classifier model.

[0244] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0245] The trained model is used to process new samples to achieve the task of vibration analysis and fault prediction of large shafts of amusement facilities. In one embodiment, the new samples to be processed are the vibration data of large shafts of amusement facilities collected from the Internet of Things platform. The data format is the same as the training samples. Let the new sample data be D new ,in N new Represents the number of new sample data. Use the trained feature extraction model to extract new samples D new Extract features from the data, let f extract () represents the feature extraction function, and the extracted features are expressed as:

[0246]

[0247] In the formula, F new is the feature set of the new sample.

[0248] Furthermore, the trained data dimensionality reduction model is used to extract the features F new Perform dimensionality reduction processing and set the data dimensionality reduction function as g reduce (), then the feature after dimensionality reduction is expressed as:

[0249]

[0250] In the formula, F reduced is the feature set after dimensionality reduction.

[0251] Furthermore, the trained classifier model is used to classify the reduced feature F reduced For classification prediction, let the classifier function be h classify (), then the prediction result for the new sample is In this embodiment, the classification categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults, a total of 5 categories.

[0252] The innovative features of the present invention mainly include the following aspects:

[0253] 1. The present invention adopts quantum state feedback to generate adversarial networks. In the process of the generator receiving quantum state feedback, the feedback contains additional information about the internal state of the discriminator. Through quantum state feedback, the generator can produce more realistic and diverse data samples.

[0254] 2. Neural network parameter optimization based on the principle of fusion phenomenon imitates the nuclear fusion process in nature to optimize the weights and biases of the neural network. The fusion process improves the stability of neural network training and avoids the common gradient disappearance or explosion problems of traditional neural networks.

[0255] 3. The autoencoder is combined with the Markov chain Monte Carlo method to optimize the training process of the autoencoder and improve the efficiency and accuracy of feature dimensionality reduction.

[0256] 4. The elastic net classification algorithm based on heterogeneous swarm optimization uses quantum bits to represent the state of the solution, and combines the optimization strategy of heterogeneous swarms to perform parameter search.

[0257] Based on the above innovations, the following technical effects are achieved:

[0258] 1. The improved parameter optimization method of the present invention solves technical challenges that may arise during the training process, such as gradient disappearance, thereby ensuring the stability and reliability of model training.

[0259] 2. By generating data expanded by adversarial networks, limited or unbalanced data can be used more effectively, reducing dependence on large-scale labeled data.

[0260] 3. The method proposed in the present invention can effectively identify and predict various types of faults, provide scientific and accurate data support for the operation and maintenance of the Ferris wheel, and help reduce faults and improve safety.

[0261] The present invention also provides an application scenario, which applies the above-mentioned method for predicting the fault of a large shaft of an amusement facility. Specifically: The method for predicting the fault of a large shaft of an amusement facility provided in this embodiment can be applied in the scenario of predicting the vibration fault of a large shaft of an amusement facility. The scenario of predicting the vibration fault of a large shaft of an amusement facility includes a data acquisition link, a feature extraction link, a dimensionality reduction link and a classification prediction link; the acquired vibration data of the large shaft of an amusement facility is subjected to feature extraction and dimensionality reduction processing and enters the classification prediction link. Specifically, in the process of processing the vibration data of the large shaft of an amusement facility, the classification category of the fault can be obtained through feature extraction, dimensionality reduction processing and classification prediction.

[0262] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store vibration data of large shafts of amusement facilities. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting faults of large shafts of amusement facilities is implemented.

[0263] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0264] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the above-mentioned method embodiments are implemented when the processor executes the computer program.

[0265] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above-mentioned method embodiments are implemented.

[0266] 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 the present invention 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.

[0267] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0268] The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0269] 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.

[0270] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, 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 the present invention.

Claims

1. A method for predicting large shaft failures of amusement facilities, characterized in that: The method for predicting failure of large shafts of amusement facilities comprises: Acquire the vibration data of the large shaft of the amusement facility; the vibration data of the large shaft of the amusement facility includes: the main shaft speed per hour, the vibration amplitude, the main shaft temperature, the vibration frequency, the acceleration, the torque, the load weight, the power consumption and the data collection time point; Extracting features from the vibration data of the large shaft of the amusement facility to obtain extracted features; Performing dimensionality reduction processing on the extracted features to obtain features after dimensionality reduction; Classification prediction is performed on the features after dimension reduction to obtain classification categories; the classification categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults.

2. The method for predicting large shaft failures of amusement facilities according to claim 1, characterized in that: Feature extraction is performed on the large shaft vibration data of the amusement facility to obtain the extracted features, which specifically include: The vibration data of the large shaft of the amusement facility is input into a feature extraction model to obtain extracted features; the feature extraction model is a neural network parameter optimization algorithm based on the principle of fusion phenomenon.

3. The method for predicting large shaft failures of amusement facilities according to claim 2, characterized in that: The method for determining the feature extraction model is: Acquire historical vibration data of large shafts of amusement facilities; the historical vibration data of large shafts of amusement facilities include: main shaft speed per hour, vibration amplitude, main shaft temperature, vibration frequency, acceleration, torque, load weight, power consumption and data collection time point; The vibration data of the large shaft of the historical amusement facility are annotated to obtain annotated data; the annotated categories include: bearing damage, load imbalance, looseness, rolling element damage and other faults; According to the labeled data, a data expansion model is used to expand the vibration data of the large axis of the historical amusement facility to obtain expanded data; the data expansion model is a quantum state feedback generative adversarial network algorithm; Initializing the parameters of the neural network according to the expanded data to obtain initialized neural network parameters; Performing a fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set; determining an energy state of the synthesized parameter set; Adjust the temperature of the nuclear reaction according to the amount of energy change; Determine whether the energy state of the synthesized parameter set reaches a dynamic equilibrium state, and obtain a first determination result; If the first judgment result is no, return to the step of "performing a fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set"; If the first judgment result is yes, then updating the parameter configuration based on energy optimization to obtain updated neural network parameters; Determine whether the energy state of the updated neural network parameters converges to obtain a second determination result; If the second judgment result is no, then return to step "perform fusion candidate check on the initialized neural network parameters to obtain a synthesized parameter set; If the second judgment result is yes, the final optimized neural network parameters are output to obtain a feature extraction model.

4. The method for predicting large shaft failures of amusement facilities according to claim 3, characterized in that: The method for determining the data expansion model is: Initializing the quantum state parameters, the initial quantum state, the network parameters of the generator, and the network parameters of the discriminator according to the labeled data to obtain the initialized quantum state parameters, the initial quantum state, the network parameters of the generator, and the network parameters of the discriminator; Generate quantum state encoding according to the initialized quantum state parameters and the initial quantum state; Generate a virtual data sample according to the quantum state encoding and input random noise; According to the virtual data sample, a discriminator is used to give a discriminant result; Adjust the quantum state parameters according to the discrimination results, the real sample labels, and the loss function of the discriminator; Use gradient descent method to update the network parameters of generator and discriminator; Determine whether the condition for stopping iteration is met, and obtain a third determination result; If the third judgment result is no, then return to step "generating a virtual data sample according to the quantum state encoding and the input random noise"; If the third judgment result is yes, the final quantum state parameters, network parameters of the generator and network parameters of the discriminator of the quantum state feedback generative adversarial network algorithm are determined to obtain a data expansion model.

5. The method for predicting large shaft failures of amusement facilities according to claim 4, characterized in that: Performing dimensionality reduction processing on the extracted features to obtain features after dimensionality reduction specifically includes: The extracted features are input into a feature dimensionality reduction model to obtain reduced features; the feature dimensionality reduction model is an autoencoder algorithm.

6. The method for predicting failure of large shafts of amusement facilities according to claim 5, characterized in that: The method for determining the feature dimension reduction model is: Initialize the parameters of the autoencoder algorithm to obtain initialized encoder parameters and decoder parameters; Generate a series of samples using the Markov chain Monte Carlo method; the samples include a number of historically extracted features; the historically extracted features are features obtained by inputting the expanded data into a feature extraction model; Inputting the sample into an autoencoder algorithm to obtain an output of the autoencoder algorithm, wherein the output of the autoencoder algorithm is reconstructed data corresponding to the sample; Determining a loss value according to the sample, the reconstructed data corresponding to the sample and the loss function; According to the loss value, the encoder parameters and the decoder parameters of the autoencoder algorithm are optimized to obtain a feature dimensionality reduction model.

7. The method for predicting failure of large shafts of amusement facilities according to claim 6, characterized in that: The features after dimension reduction are classified and predicted to obtain classification categories, which specifically include: The reduced-dimensional features are input into a classifier model to obtain classification categories; the classifier model is an elastic net classification algorithm based on heterogeneous bee colony optimization.

8. The method for predicting failure of large shafts of amusement facilities according to claim 7, characterized in that: The method for determining the classifier model is: Initialize the quantum bit state, position and speed of each swarm member according to the features after historical dimension reduction, and obtain the initialized quantum bit state, position and speed of each swarm member; the features after historical dimension reduction are the features obtained by inputting the features after historical extraction into the feature dimension reduction model; Determine the performance index of each swarm member according to the position of each swarm member after initialization; Determine the location of the swarm member with the best performance index; Based on the position of the swarm member with the best performance index, a heterogeneous swarm optimization strategy is adopted to update the qubit state, position and speed of each swarm member; Determine whether the condition for stopping iteration is met, and obtain a fourth determination result; If the fourth judgment result is no, then return to step "determining the performance index of each swarm member according to the position of each swarm member after initialization"; If the fourth judgment result is yes, the position of the swarm member with the best performance index is determined as the optimal model parameter to obtain the classifier model.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting large shaft faults of an amusement facility according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting large shaft failures of an amusement facility according to any one of claims 1 to 8 is implemented.

Images