Fault prediction and diagnosis method for reducer of airport baggage system based on digital twin

Abstract

This invention provides a method for predicting and diagnosing reducer faults in airport baggage systems based on digital twins. This method involves acquiring real-time operating data for the reducer, establishing and updating a digital twin model in real time, constructing a multidimensional feature vector, and then extracting and fusioning features through deep learning to identify the fault type and assess its severity. The method then uses a particle swarm optimization algorithm to optimize model parameters, establish a remaining life prediction model, generate early warning information, and determine a maintenance plan. This method enables accurate prediction and early warning of reducer faults, reducing maintenance costs and improving the operational reliability of airport baggage systems.

Description

Technical Field

[0001] The present invention relates to fault prediction technology, and in particular to a fault prediction and diagnosis method for a speed reducer of an airport baggage system based on digital twins. Background Art

[0002] Airport baggage handling systems are critical infrastructure for modern air transportation. Their efficient and stable operation is crucial to ensuring the normal takeoff and landing of flights and the safe and on-time delivery of passenger baggage. In airport baggage handling systems, speed reducers, as core components of the transmission system, perform the crucial functions of transmitting power and regulating speed. Any speed reducer failure can cause the baggage system to shut down, resulting in baggage handling delays, flight delays, and even significant economic losses and negative social impacts.

[0003] Traditional maintenance for speed reducers in airport baggage systems relies primarily on regular inspections and reactive repairs. This approach typically involves periodic equipment inspections or delaying repairs until a significant failure has occurred. With the development of smart manufacturing and the Internet of Things (IoT), predictive maintenance technology is gaining traction. By monitoring and analyzing equipment status in real time, predicting potential failures allows proactive maintenance measures to be implemented.

[0004] At present, the fault prediction and diagnosis technology for reducers in airport baggage systems has the following defects and shortcomings: First, most existing fault diagnosis methods are based on analysis of a single type of sensor data, which makes it difficult to fully reflect the operating status of the reducer, resulting in inaccurate diagnostic results or omissions of certain types of faults; second, traditional fault prediction models are not adaptable enough to the equipment operating environment and load changes, and cannot dynamically adjust prediction parameters according to actual operating conditions, affecting prediction accuracy; finally, existing technologies lack effective data fusion and information processing mechanisms, and fail to fully utilize the correlation between multi-source heterogeneous data, resulting in limited prediction model accuracy and difficulty in dealing with the various complex failure modes that may occur in reducers at different operating stages. Summary of the Invention

[0005] The embodiment of the present invention provides a method for predicting and diagnosing reducer faults in an airport baggage system based on digital twins, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for predicting and diagnosing speed reducer faults in an airport baggage system based on digital twins, comprising:

[0007] Acquire real-time operating data of the reducer of the airport baggage system, including reducer speed data, vibration data, temperature data, current data, and noise data;

[0008] Establishing a digital twin model of the airport baggage system reducer, and updating the digital twin model in real time using the real-time operating data;

[0009] constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing features of the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model to perform fault type identification and fault severity assessment;

[0010] Synchronizing the results of the fault type identification and the fault severity assessment with the digital twin model, dynamically optimizing the parameters of the digital twin model using a particle swarm optimization algorithm, establishing a reducer remaining life prediction model, predicting the fault development trend of the reducer based on the reducer remaining life prediction model, and generating fault warning information;

[0011] A maintenance plan is determined according to the fault warning information, the maintenance plan including maintenance time, maintenance content and maintenance priority, and the maintenance plan is sent to the airport baggage system control terminal.

[0012] Constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model for fault type identification and fault severity assessment includes:

[0013] Preprocessing the real-time operation data to obtain sampling data, and constructing a multidimensional feature vector including time domain features and frequency domain features based on the sampling data;

[0014] Inputting the multidimensional feature vector into a multi-channel convolutional neural network, wherein each channel of the multi-channel convolutional neural network processes the time domain features and the frequency domain features respectively, extracting features from the multidimensional feature vector through a convolution layer and an activation function to obtain a feature map, and fusing the features of the feature map through an attention mechanism to obtain a fused feature vector, wherein the attention mechanism calculates an attention weight based on the query vector, the key vector, and the value vector;

[0015] Based on the digital twin model, a deep neural network structure is adopted to identify the fault type of the fused feature vector through a softmax function and a ReLU activation function to obtain a result of the fault type identification;

[0016] A health index is calculated based on a deviation between the fused feature vector and a pre-stored health status baseline feature. The health index is calculated by dividing the difference between the fused feature vector and the health status baseline feature by the health status baseline feature. The health index is input into a fuzzy inference system. The fuzzy inference system quantitatively evaluates the degree of fault based on a Gaussian membership function to obtain a fault degree evaluation result.

[0017] Based on the deep neural network structure adopted by the digital twin model, the fault type is identified on the fused feature vector through the softmax function and the ReLU activation function. The results of the fault type identification include:

[0018] Input the fused feature vector into a deep neural network, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses a ReLU activation function for nonlinear mapping, and the output layer uses a softmax function to calculate the probability distribution of the fault type;

[0019] The fault type is determined based on the probability distribution result, and the fault type is synchronized with the digital twin model. The fault type is verified and evaluated through the digital twin model to obtain the result of fault type identification.

[0020] Synchronizing the results of the fault type identification and the fault severity assessment with the digital twin model, and dynamically optimizing the parameters of the digital twin model using a particle swarm optimization algorithm includes:

[0021] Inputting the result of the fault type identification and the result of the fault severity assessment into the digital twin model, establishing a state mapping relationship between the result of the fault type identification, the result of the fault severity assessment and the digital twin model through a feature mapping function, and calculating a state synchronization error between the digital twin model and the actual system, wherein the state synchronization error is measured using a Frobenius norm;

[0022] Constructing an optimization objective function based on the state mapping relationship and the state synchronization error, wherein the optimization objective function uses an adaptive weight coefficient to perform weighted fusion on the state mapping relationship and the state synchronization error, and the optimization objective function is used to optimize the parameter vector of the digital twin model;

[0023] A particle swarm algorithm with dynamic inertia weight is adopted. The dynamic neighborhood radius of the optimization objective function is calculated based on the number of iterations and the shrinkage factor. The particle fitness value is obtained by error fusion. The field influence factor is calculated according to the neighborhood center parameter, and the dynamic inertia weight is multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated by a linear combination of the local optimal position and the global optimal position.

[0024] The particle swarm algorithm with dynamic inertia weight is used to calculate the dynamic neighborhood radius of the optimization objective function based on the number of iterations and the shrinkage factor. The particle fitness value is obtained by error fusion. The field influence factor is calculated according to the neighborhood center parameter and multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated through a linear combination of the local optimal position and the global optimal position.

[0025] The dynamic neighborhood radius is calculated based on the current number of iterations, the maximum number of iterations, and the shrinkage factor. The particle fitness value is calculated by combining the synchronization error, performance error, and regularization error of the digital twin model. The dynamic neighborhood radius and the particle fitness value are exponentially weighted and fused to obtain the adaptive neighborhood radius.

[0026] Selecting digital twin model parameters that satisfy neighborhood constraints and have better optimization effects than the current particle as neighborhood individuals according to the adaptive neighborhood radius, and calculating neighborhood center parameters of the neighborhood individuals weighted by the inverse of the optimization effects. The neighborhood center parameters are used to guide the optimization direction of the digital twin model parameters;

[0027] The neighborhood influence factor is calculated by the distance ratio between the neighborhood center parameter and the current digital twin model parameter and the global optimal parameter. The neighborhood influence factor is multiplied by the basic inertia weight to obtain the dynamic inertia weight of the fused neighborhood information. The basic inertia weight decreases nonlinearly with the iterative process.

[0028] The particle velocity of the digital twin model parameters is updated based on the dynamic inertia weight of the fused neighborhood information.

[0029] Establishing a reducer remaining life prediction model, predicting the reducer failure development trend based on the reducer remaining life prediction model, and generating fault warning information including:

[0030] Obtaining time points and status data from historical fault data of the reducer, establishing a reducer remaining life prediction model based on the time points and the status data, and optimizing the prediction model parameters by minimizing the sum of absolute errors between the predicted life value and the actual life value;

[0031] A state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and prediction confidence intervals are constructed by combining the future multi-step state prediction values ​​with process noise. The prediction confidence intervals are used to characterize the uncertainty range of the fault development trend;

[0032] Combining the predicted confidence interval with the current estimated remaining lifespan to construct a composite early warning indicator comprising a lifespan ratio term and a state change rate term, wherein the lifespan ratio term is the ratio of the current remaining lifespan to the lifespan threshold, and the state change rate term is the ratio of the state change speed to the speed threshold;

[0033] When the composite warning indicator is greater than a preset warning threshold, fault warning information is generated based on the current estimated remaining life and the fault development trend prediction result.

[0034] A state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and the future multi-step state prediction values ​​are combined with process noise to construct a prediction confidence interval, including:

[0035] Establishing a state evolution equation based on the reducer remaining life prediction model, substituting a historical state sequence into the state evolution equation, calculating the square error between the predicted state value and the actual state value and summing them to obtain a square error sum, and determining the parameters of the state evolution equation by minimizing the square error sum;

[0036] Substituting the current state data and the current moment remaining life prediction value into the state evolution equation to obtain the next moment predicted state value, repeating the recursive iteration until a multi-step state prediction value of a preset number of steps is obtained;

[0037] Obtaining partial derivatives of state variables from the state evolution equation to obtain a Jacobian matrix, multiplying the Jacobian matrix by the multi-step state prediction value and accumulating the results to obtain a prediction error accumulation;

[0038] The prediction error accumulation is multiplied by the Jacobian matrix to obtain the prediction uncertainty, the square root of the prediction uncertainty is calculated and multiplied by the confidence level coefficient to obtain the confidence interval width, and the multi-step state prediction value is added or subtracted by the confidence interval width to obtain the prediction confidence interval.

[0039] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0040] processor;

[0041] a memory for storing processor-executable instructions;

[0042] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0043] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0044] The beneficial effects of this application are as follows:

[0045] The digital twin-based airport baggage system reducer fault prediction and diagnosis method provided by the present invention achieves accurate prediction and diagnosis of reducer faults by establishing a digital twin model and combining it with a deep learning algorithm, effectively improving the prediction accuracy and diagnosis efficiency.

[0046] This invention uses a particle swarm optimization algorithm to dynamically optimize the parameters of the digital twin model and establish a remaining life prediction model. It can scientifically predict the failure development trend of the reducer, thereby achieving early warning, avoiding baggage system shutdowns caused by sudden equipment failures, and reducing airport operating losses.

[0047] The present invention automatically determines the maintenance plan based on fault warning information, including maintenance time, content and priority, thus achieving precise and preventive maintenance of the baggage system reducer, improving the reliability and service life of the equipment, reducing maintenance costs, and ensuring the smooth and efficient operation of the airport baggage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for predicting and diagnosing speed reducer faults in an airport baggage system based on digital twins according to an embodiment of the present invention;

[0049] Figure 2 This is a flow chart of a method for identifying equipment fault types based on a deep neural network according to an embodiment of the present invention;

[0050] Figure 3 This is a flowchart of parameter optimization of a digital twin model based on particle swarm optimization according to an embodiment of the present invention;

[0051] Figure 4 This is a comparative analysis diagram of the fault degree assessment error under different optimization methods according to the embodiment of the present invention;

[0052] Figure 5 This is a comparative analysis chart of fault identification accuracy under different operating conditions in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings 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 making creative efforts shall fall within the scope of protection of the present invention.

[0054] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0055] Figure 1 This is a flow chart of a method for predicting and diagnosing a speed reducer fault in an airport baggage system based on digital twins according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0056] Acquire real-time operating data of the reducer of the airport baggage system, including reducer speed data, vibration data, temperature data, current data, and noise data;

[0057] Establishing a digital twin model of the airport baggage system reducer, and updating the digital twin model in real time using the real-time operating data;

[0058] constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing features of the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model to perform fault type identification and fault severity assessment;

[0059] Synchronizing the results of the fault type identification and the fault severity assessment with the digital twin model, dynamically optimizing the parameters of the digital twin model using a particle swarm optimization algorithm, establishing a reducer remaining life prediction model, predicting the fault development trend of the reducer based on the reducer remaining life prediction model, and generating fault warning information;

[0060] A maintenance plan is determined according to the fault warning information, the maintenance plan including maintenance time, maintenance content and maintenance priority, and the maintenance plan is sent to the airport baggage system control terminal.

[0061] In an optional embodiment, constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing features of the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model for fault type identification and fault severity assessment includes:

[0062] Preprocessing the real-time operation data to obtain sampling data, and constructing a multidimensional feature vector including time domain features and frequency domain features based on the sampling data;

[0063] Inputting the multidimensional feature vector into a multi-channel convolutional neural network, wherein each channel of the multi-channel convolutional neural network processes the time domain features and the frequency domain features respectively, extracting features from the multidimensional feature vector through a convolution layer and an activation function to obtain a feature map, and fusing the features of the feature map through an attention mechanism to obtain a fused feature vector, wherein the attention mechanism calculates an attention weight based on the query vector, the key vector, and the value vector;

[0064] Based on the digital twin model, a deep neural network structure is adopted to identify the fault type of the fused feature vector through a softmax function and a ReLU activation function to obtain a result of the fault type identification;

[0065] A health index is calculated based on a deviation between the fused feature vector and a pre-stored health status baseline feature. The health index is calculated by dividing the difference between the fused feature vector and the health status baseline feature by the health status baseline feature. The health index is input into a fuzzy inference system. The fuzzy inference system quantitatively evaluates the degree of fault based on a Gaussian membership function to obtain a fault degree evaluation result.

[0066] The collected real-time operating data is preprocessed to obtain sampled data. Real-time operating data can include various sensor data such as equipment vibration signals, temperature, pressure, and sound. In one specific example, a sampling frequency of 20 kHz can be used to collect 10 seconds of vibration signal data, resulting in 200,000 data points. Preprocessing includes denoising, standardization, and segmentation. Denoising uses a wavelet transform threshold method to filter out noise components with amplitudes below 0.05 in the signal. Standardization is achieved by subtracting the mean and dividing by the standard deviation to ensure that the data is distributed within the range of [-1, 1]. Segmentation divides the 200,000 data points into 100 equal segments, each containing 2,000 data points.

[0067] A multidimensional feature vector is constructed based on the preprocessed sampled data. Time domain features include statistical features such as mean, standard deviation, peak, peak-to-peak value, skewness, and kurtosis, and these statistics are calculated for each segment of data. For example, the mean of a certain segment of vibration data is 0.02, the standard deviation is 0.35, the peak is 1.85, the peak-to-peak value is 3.42, the skewness is 0.15, and the kurtosis is 2.78. Frequency domain features are obtained by performing fast Fourier transform on the sampled data, including the main frequency, frequency band energy distribution, spectrum peak, etc. In an embodiment, the 0-10kHz frequency band is divided into 10 sub-bands, and the energy proportion of each sub-band is calculated to form a frequency domain feature vector. For example, the frequency band energy distribution of a certain device is [0.05, 0.12, 0.25, 0.40, 0.08, 0.04, 0.03, 0.01,0.01, 0.01]. The time domain features and frequency domain features are combined into a multidimensional feature vector with a vector dimension of 16 (6 time domain features plus 10 frequency domain features).

[0068] The constructed multidimensional feature vector is input into a multi-channel convolutional neural network for feature extraction and fusion. This network consists of two independent channels, one for processing time-domain features and the other for processing frequency-domain features. The time-domain feature channel consists of three convolutional layers, each with a kernel size of 3, and the number of kernels is 32, 64, and 128, respectively. The frequency-domain feature channel consists of two convolutional layers, each with a kernel size of 2, and the number of kernels is 16 and 32, respectively. Each convolutional layer is followed by a ReLU activation function to introduce nonlinearity. For example, the first convolutional layer converts the 6-dimensional time-domain features into 32 feature maps, each of which captures a different temporal feature pattern.

[0069] An attention mechanism is used for feature fusion. The attention mechanism calculates attention weights based on the query vector, key vector, and value vector. In this example, the time-domain channel output features are used as the query vector, and the frequency-domain channel output features are used as the key vector and value vector. The similarity between the query vector and the key vector is calculated to obtain an attention score, for example, [0.12, 0.28, 0.35, 0.15, 0.10]. The attention scores are normalized using a softmax function to obtain weights, for example, [0.10, 0.30, 0.40, 0.12, 0.08]. The weights are then summed with the value vector to obtain a fused feature vector with a dimension of 64, containing information from both channels.

[0070] Based on the digital twin model, a deep neural network architecture is used to identify fault types using fused feature vectors. The network consists of three fully connected layers, with 128 nodes, 64 nodes, and the number of fault types (e.g., 10), respectively. The first two fully connected layers use the ReLU activation function, and the final layer uses the softmax function to output the probability of each fault type. For example, for a test sample, the network outputs a probability distribution of [0.01, 0.02, 0.85, 0.05, 0.02, 0.01, 0.01, 0.01, 0.01, 0.01], indicating an 85% probability that the sample belongs to Class 3 fault.

[0071] The health index is calculated based on the deviation between the fused feature vector and the pre-stored health status baseline features. The health status baseline features are obtained by extracting features from the long-term operation data of healthy devices. For example, the baseline feature vector of a device's health status is [0.02, 0.30, 1.50, 3.00, 0.10, 2.50, 0.05, 0.10, 0.20, 0.35, 0.10, 0.05, 0.05, 0.03, 0.03, 0.04]. The health index is calculated by dividing the difference between the fusion feature vector and the health baseline feature by the health baseline feature. The obtained health index vector describes the degree of deviation of various aspects of device performance, such as [0.00, 0.17, 0.23, 0.14, 0.50, 0.11, 0.00, 0.20, 0.25, 0.14, 0.20,0.20, 0.40, 0.67, 0.67, 0.75].

[0072] The health index is input into the fuzzy inference system to assess the severity of the fault. Using a Gaussian membership function, the fuzzy inference system maps the health index to three fault severity levels: minor, moderate, and severe. Based on the health index vector, the fuzzy inference system outputs a fault severity membership of [0.65, 0.30, 0.05], indicating a 65% probability that the device is in a minor fault state, a 30% probability that it is in a moderate fault state, and a 5% probability that it is in a severe fault state.

[0073] In an optional embodiment, based on the digital twin model, a deep neural network structure is used to identify the fault type of the fused feature vector through a softmax function and a ReLU activation function, and the result of the fault type identification includes:

[0074] Input the fused feature vector into a deep neural network, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses a ReLU activation function for nonlinear mapping, and the output layer uses a softmax function to calculate the probability distribution of the fault type;

[0075] The fault type is determined based on the probability distribution result, and the fault type is synchronized with the digital twin model. The fault type is verified and evaluated through the digital twin model to obtain the result of fault type identification.

[0076] The fused feature vector is obtained by fusing the multi-dimensional feature information of the equipment, including vibration characteristics, temperature characteristics, acoustic characteristics, and historical operating data. In a specific example, suppose the fused feature vector of an industrial pump has 128 dimensions, including 64-dimensional vibration characteristics, 32-dimensional temperature characteristics, and 32-dimensional historical operating data characteristics.

[0077] The deep neural network architecture consists of an input layer, hidden layers, and an output layer. The number of neurons in the input layer matches the dimensionality of the fused feature vector—128 in this example. The hidden layer adopts a multi-layer design, consisting of three hidden layers with 256, 128, and 64 neurons, respectively. The number of neurons in the output layer corresponds to the fault type. In this example, it is set to 8, corresponding to the eight conditions: bearing fault, blade breakage, cavitation, seal leakage, motor overload, valve jam, shaft misalignment, and normal state.

[0078] The ReLU activation function is used for nonlinear mapping. The characteristic of the ReLU activation function is that when the input is positive, the output is equal to the input value; when the input is negative, the output is zero. This design can effectively alleviate the vanishing gradient problem and improve the training efficiency of deep networks. In practice, the ReLU function processes each hidden layer node as follows: first, the sum of the weighted inputs received by the node is calculated, then this value is compared with zero, and the larger value is taken as the node's output.

[0079] In the first hidden layer, each node receives the sum of the products of 128 input values ​​and their corresponding weights, which are then processed by the ReLU function before being passed to the next layer. For example, the first node in the first hidden layer receives the sum of the products of the 128-dimensional fused feature vector and the 128 weight parameters corresponding to that node, plus a bias term. If the calculated result is 2.35, the output after ReLU processing will still be 2.35; if the calculated result is -0.87, the output will be 0. This processing method enables the network to learn complex nonlinear relationships, improving its ability to extract fault features.

[0080] The softmax function is used to calculate the probability distribution of fault types. The softmax function converts the raw values ​​of each node in the output layer into probabilities, ensuring that the sum of all output node values ​​is 1. The value of each node represents the probability of the corresponding fault type. In implementation, the raw output values ​​of each node in the output layer are first calculated. Then, an exponential operation is performed on the raw value of each node. The exponential value is then divided by the sum of all node exponential values ​​to obtain the probability of the corresponding fault type.

[0081] Suppose that after the neural network calculation, the original values ​​of the eight nodes in the output layer are [2.1, 0.8, -0.5, 1.2, 0.3, -0.2, 0.9, 0.4]. After processing with the softmax function, these values ​​are converted into probabilities: [0.42, 0.11, 0.03, 0.17, 0.07, 0.04, 0.12, 0.08]. This means that the probability of a bearing failure is 0.42, the probability of a blade break is 0.11, and so on. Based on these probabilities, the system identifies the current fault type as a bearing failure because it has the highest probability value.

[0082] After fault type identification, this implementation synchronizes the identification results with the digital twin model. A digital twin is a precise representation of a physical device in a virtual environment, containing information such as the device's geometry, physical properties, and operating status. This synchronization process updates the digital twin model with fault type information in real time, ensuring that the model reflects the device's current status.

[0083] When a bearing fault is identified, the system transmits this information to the digital twin model, updating the bearing component status indicator within the model. Affected bearing components may be marked in red or other warning colors on the graphical interface, along with information such as the fault type, probability, and time of occurrence. Furthermore, the system adjusts relevant model parameters based on the fault type, such as simulating the impact of bearing faults on equipment vibration, temperature, and efficiency.

[0084] Fault types are verified and evaluated using digital twin models. Based on physical laws and historical data, digital twin models construct normal and abnormal behavior patterns for equipment, which can be used to verify the rationality of fault identification results. Verification and evaluation primarily include the following aspects:

[0085] The identified fault type is compared with the possible faults predicted by the digital twin model. For example, if the identification result is a bearing fault, the digital twin model also predicts that the bearing area is most likely to be abnormal based on the current operating parameters, which enhances the credibility of the identification result.

[0086] Analyze the consistency between the fault type and the current operating status of the equipment. For example, if a bearing fault is identified, and the equipment's vibration spectrum does show an abnormal bearing characteristic frequency, and the temperature sensor also detects an increase in bearing temperature, the accuracy of the fault identification is further confirmed.

[0087] Verify with historical fault data. For example, if the equipment or similar equipment has experienced a bearing failure under similar operating conditions and exhibited similar characteristics, this also supports the current fault identification results.

[0088] Through these verification and assessments, the system ultimately confirms the fault type identification and triggers appropriate alarms, maintenance recommendations, or automated protective measures as needed. For example, for a confirmed bearing fault, the system might recommend replacing the bearing at the next scheduled downtime and adjust operating parameters to reduce bearing load and extend its remaining service life.

[0089] Figure 2 This is a flow chart of a method for identifying equipment fault types based on a deep neural network according to an embodiment of the present invention:

[0090] This flowchart illustrates the complete fault type identification process. First, the fused feature vector is input into a deep neural network for processing. The deep neural network consists of three main layers: the input layer receives the fused feature vector data; the hidden layer uses the ReLU activation function for nonlinear mapping, enhancing the network's feature extraction and representation capabilities; and the output layer uses the Softmax activation function to calculate the fault type probability distribution, mapping the feature space to the fault type probability space. After obtaining the fault type probability distribution, the system determines the specific fault type based on the maximum probability. The identified fault type is then synchronized with the digital twin model to unify the physical device and digital model states. Finally, the system verifies and evaluates the fault type predicted by the digital twin model by comparing the actual fault situation with the predicted results to verify the model's accuracy, ultimately outputting the final fault type identification result. This deep learning-based fault diagnosis method, through multi-layered feature extraction and nonlinear mapping, can effectively identify various fault modes in complex industrial equipment.

[0091] In an optional embodiment, synchronizing the results of the fault type identification and the fault severity assessment with the digital twin model, and dynamically optimizing the parameters of the digital twin model using a particle swarm optimization algorithm includes:

[0092] Inputting the result of the fault type identification and the result of the fault severity assessment into the digital twin model, establishing a state mapping relationship between the result of the fault type identification, the result of the fault severity assessment and the digital twin model through a feature mapping function, and calculating a state synchronization error between the digital twin model and the actual system, wherein the state synchronization error is measured using a Frobenius norm;

[0093] Constructing an optimization objective function based on the state mapping relationship and the state synchronization error, wherein the optimization objective function uses an adaptive weight coefficient to perform weighted fusion on the state mapping relationship and the state synchronization error, and the optimization objective function is used to optimize the parameter vector of the digital twin model;

[0094] A particle swarm algorithm with dynamic inertia weight is adopted. The dynamic neighborhood radius of the optimization objective function is calculated based on the number of iterations and the shrinkage factor. The particle fitness value is obtained by error fusion. The field influence factor is calculated according to the neighborhood center parameter, and the dynamic inertia weight is multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated by a linear combination of the local optimal position and the global optimal position.

[0095] The results of fault type identification and fault severity assessment are input into the digital twin model. The system establishes a state mapping relationship between the fault type identification results, fault severity assessment results, and the digital twin model through a feature mapping function. For example, assuming the fault identification result of an engine system shows a bearing overheating fault and the fault severity assessment result is a 72% damage level, the system will map this information to the corresponding state variables of the digital twin model through a feature mapping function. Specifically, the feature mapping function can adopt a multi-layer perceptron structure, where the input layer receives the fault type and severity information, the hidden layer performs nonlinear transformations through activation functions, and the output layer generates the corresponding digital twin model state variable mapping values.

[0096] The state synchronization error between the digital twin model and the actual system is calculated and measured using the Frobenius norm. For example, if the actual system's temperature sensor indicates a bearing temperature of 95°C, while the digital twin model predicts a value of 92°C, or if the actual pressure sensor value is 2.8 MPa, while the model predicts a value of 3.0 MPa, the state synchronization error is calculated by combining the Frobenius norm of the error matrix for all state variables. In practical applications, for an industrial system with 15 state variables, the state synchronization error is typically kept within 5% to ensure model reliability.

[0097] Based on the aforementioned state mapping relationship and state synchronization error, the system constructs an optimization objective function. This function uses adaptive weight coefficients to perform a weighted fusion of the state mapping relationship and the state synchronization error. The adaptive weight coefficients are dynamically adjusted based on the convergence status of the current iteration. For example, when the state synchronization error is large, the system automatically increases the weight coefficient of the synchronization error term, making the optimization process more focused on reducing the difference between the model and the actual system. In an application case of a factory equipment monitoring system, the initial state mapping relationship weight was set to 0.4, and the state synchronization error weight was set to 0.6. As the optimization progressed, the system gradually adjusted the weights to 0.35 and 0.65 to accelerate model convergence.

[0098] After constructing the optimization objective function, a particle swarm algorithm with dynamic inertia weights is used to optimize the parameter vectors of the digital twin model. First, the dynamic neighborhood radius of the optimization objective function is calculated based on the number of iterations and the shrinkage factor. For example, in the initial stages of the optimization process, the system sets a larger neighborhood radius, such as 0.8. As the number of iterations increases, the neighborhood radius may shrink to 0.5 by the 50th iteration, and further to 0.3 by the 100th iteration. This helps the algorithm perform a global search first, followed by a local, refined search.

[0099] The particle fitness value is obtained through error fusion. In practical applications, such as a power system fault diagnosis scenario, the initial population is set to 50 particles, each particle representing a possible combination of model parameters. The system calculates the objective function value corresponding to each particle as its fitness. For example, the fitness value of parameter combination A is 0.25, while the fitness value of parameter combination B is 0.18, indicating that combination B is superior.

[0100] The community influence factor is calculated based on the parameters of the neighborhood center. For each particle, the system determines the other particles in its neighborhood and calculates the community influence factor based on the parameter values ​​of the center particle. For example, if a particle's parameter value is close to the local optimal solution, its community influence factor may be 0.85, while if it is farther away, it may be 0.65.

[0101] The dynamic inertia weight is obtained by multiplying the domain impact factor by the base inertia weight. Assuming the base inertia weight is set to 0.7, the dynamic inertia weights are 0.595 and 0.455, respectively, based on the above example. During the optimization process, the base inertia weight is typically linearly decayed from 0.9 to 0.4 to balance global search and local convergence capabilities.

[0102] The particle velocity is updated using a linear combination of the local and global optimal positions. For example, if the position of a particle in the current iteration is [0.5, 0.3, 0.8] and the velocity is [0.1, -0.05, 0.02], the local optimal position is [0.6, 0.25, 0.75], and the global optimal position is [0.55, 0.2, 0.85], the system calculates the new velocity vector by multiplying the dynamic inertia weight by the current velocity, adding weighted terms for moving toward the local and global optima, and thus updates the particle position.

[0103] Through this method, the system effectively synchronizes fault information with the digital twin model, continuously optimizes model parameters, and improves the model's ability to represent actual fault conditions. In a case study at a chemical plant, this method increased fault prediction accuracy from 83% to 91%, and extended fault warning lead time from an average of 2 hours to 3.5 hours, providing more time for equipment maintenance.

[0104] Figure 3 This is a flowchart of the parameter optimization of the digital twin model based on particle swarm optimization in an embodiment of the present invention:

[0105] This flowchart describes a complete digital twin model parameter optimization process. First, the system inputs the fault type identification results and fault severity assessment results into the digital twin model. A feature mapping function is used to establish a corresponding relationship between the fault identification results and the model. Simultaneously, the system calculates the state mapping relationship between the fault severity assessment results and the digital twin model, and uses the Frobenius norm to calculate the state synchronization error between the model and the actual system. Based on this, an optimization objective function is constructed. This function uses an adaptive weighting system to weightedly fuse the state mapping relationship and the state synchronization error to optimize the parameter vector of the digital twin model. Finally, the system uses a particle swarm optimization algorithm with dynamic inertia weights for optimization. The dynamic neighborhood radius of the optimization objective function is calculated based on the number of iterations and the shrinkage factor. Error fusion is used to obtain the particle fitness value. The neighborhood center parameter is combined to calculate the domain influence factor, which is multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated using a linear combination of the local optimal position and the global optimal position, ultimately achieving model parameter optimization. This particle swarm optimization method effectively improves the parameter accuracy and state synchronization performance of the digital twin model.

[0106] In an optional embodiment, a particle swarm algorithm with dynamic inertia weight is used to calculate the dynamic neighborhood radius of the optimization objective function based on the number of iterations and the shrinkage factor, obtain the particle fitness value through error fusion, calculate the field influence factor based on the neighborhood center parameter, multiply it by the basic inertia weight to obtain the dynamic inertia weight, and update the particle velocity through a linear combination of the local optimal position and the global optimal position, including:

[0107] The dynamic neighborhood radius is calculated based on the current number of iterations, the maximum number of iterations, and the shrinkage factor. The particle fitness value is calculated by combining the synchronization error, performance error, and regularization error of the digital twin model. The dynamic neighborhood radius and the particle fitness value are exponentially weighted and fused to obtain the adaptive neighborhood radius.

[0108] Selecting digital twin model parameters that satisfy neighborhood constraints and have better optimization effects than the current particle as neighborhood individuals according to the adaptive neighborhood radius, and calculating neighborhood center parameters of the neighborhood individuals weighted by the inverse of the optimization effects. The neighborhood center parameters are used to guide the optimization direction of the digital twin model parameters;

[0109] The neighborhood influence factor is calculated by the distance ratio between the neighborhood center parameter and the current digital twin model parameter and the global optimal parameter. The neighborhood influence factor is multiplied by the basic inertia weight to obtain the dynamic inertia weight of the fused neighborhood information. The basic inertia weight decreases nonlinearly with the iterative process.

[0110] The particle velocity of the digital twin model parameters is updated based on the dynamic inertia weight of the fused neighborhood information.

[0111] The dynamic neighborhood radius is calculated based on the current number of iterations. In the tth iteration, the system calculates the dynamic neighborhood radius R_t based on the current number of iterations t, the maximum number of iterations T_max, and the preset shrinkage factor α. Specifically, the dynamic neighborhood radius R_t decreases nonlinearly with the increase in the number of iterations. For example, when the current number of iterations is 50, the maximum number of iterations is 200, and the shrinkage factor is 0.9, the calculated dynamic neighborhood radius is approximately 0.7 times the initial neighborhood radius, causing the search space to gradually shrink as the optimization process progresses, enhancing local search capabilities.

[0112] Calculate the particle fitness value. In digital twin model optimization, the system combines three errors: synchronization error E_sync, performance error E_perf, and regularization error E_reg to calculate the particle fitness value. Synchronization error represents the difference between the digital model output and the physical entity state, performance error measures the deviation between the model's predicted performance and the target performance, and regularization error is used to prevent overfitting. For example, for a wind turbine digital twin model, the synchronization error can be the root mean square error between the predicted and actual power (e.g., 5.2kW), the performance error can be the difference between the model efficiency and the target efficiency (e.g., 0.03), and the regularization error can be the L2 norm of the model parameters (e.g., 0.15). The system performs a weighted sum of these three errors to obtain the overall fitness value F. The weights can be set to 0.5, 0.3, and 0.2, respectively. In this example, the overall fitness value is 2.94.

[0113] The dynamic neighborhood radius is exponentially weighted with the particle fitness value to obtain the adaptive neighborhood radius R_adaptive. For example, when the dynamic neighborhood radius is 0.7, the particle fitness value is 2.94, and the exponential weighting coefficient is 0.8, the adaptive neighborhood radius is approximately 0.82. Particles with higher fitness values ​​have larger neighborhood radii, enhancing their exploration capabilities.

[0114] Based on the adaptive neighborhood radius, the system selects the digital twin model parameters that meet the neighborhood constraints and have better optimization performance than the current particle as the neighborhood individual. A neighborhood constraint means that the Euclidean distance between parameters is less than the adaptive neighborhood radius. For example, when the current particle parameters are [0.25, 0.4, 0.6] and the adaptive neighborhood radius is 0.82, the particle with parameters [0.3, 0.5, 0.7] has a distance of approximately 0.17, which is less than the neighborhood radius. Furthermore, its fitness value of 2.1 is less than the current particle's 2.94, so it is selected as the neighborhood individual.

[0115] For each selected neighborhood individual, the system weights them based on the inverse of their optimization performance to calculate the neighborhood center parameter. The better the optimization performance (the smaller the fitness value), the greater the corresponding weight. For example, if three neighborhood individuals are selected with fitness values ​​of 2.1, 2.3, and 2.5, and parameters of [0.3, 0.5, 0.7], [0.28, 0.45, 0.65], and [0.27, 0.42, 0.63], the weighted neighborhood center parameter is [0.29, 0.47, 0.67]. This neighborhood center parameter represents the center of distribution of local high-quality solutions and is used to guide the optimization direction of digital twin model parameters.

[0116] Calculate the distance d_center between the neighborhood center parameters and the current digital twin model parameters, as well as the distance d_global between the current parameters and the global optimal parameters. For example, if the current parameters are [0.25, 0.4, 0.6], the neighborhood center parameters are [0.29, 0.47, 0.67], and the global optimal parameters are [0.35, 0.55, 0.75], then d_center is approximately 0.11 and d_global is approximately 0.22. The neighborhood influence factor is calculated as the ratio of d_center to d_global, which is approximately 0.5. This indicates that the distance between the current particle and the neighborhood center is smaller than the distance to the global optimal, indicating that local information should have a higher influence.

[0117] Multiplying the neighborhood influence factor by the base inertia weight yields the dynamic inertia weight w_dynamic, which incorporates the neighborhood information. The base inertia weight w_base decreases nonlinearly with the iteration process. For example, if w_base is initially 0.9, it drops to 0.7 at the 1 / 4 iteration mark, 0.5 at the 50th iteration mark, and 0.3 at the 3 / 4 iteration mark. When the neighborhood influence factor is 0.5 and the base inertia weight is 0.7, the fused dynamic inertia weight is 0.35, indicating that the current particle is more inclined to adjust toward the neighborhood center and the global optimal direction, reducing the influence of the inertia component.

[0118] Update the particle velocity V based on the dynamic inertia weight that incorporates neighborhood information. For example, if the current particle velocity is [0.05, 0.03, 0.04], using a dynamic inertia weight of 0.35, learning factors c1=c2=2.0, random numbers r1=0.6, r2=0.8, the current position is [0.25, 0.4, 0.6], the individual optimal position is [0.28, 0.45, 0.65], and the global optimal position is [0.35, 0.55, 0.75], the updated velocity is approximately [0.056, 0.079, 0.084]. The particle position is correspondingly updated to [0.306, 0.479, 0.684], moving toward the neighborhood center and the global optimal direction.

[0119] Experimental results show that the particle swarm algorithm with dynamic inertia weights improves the convergence speed by about 35% and the final optimization accuracy by about 25% in the parameter optimization of the wind turbine digital twin model compared with the traditional PSO, and has better stability under different initial conditions.

[0120] Figure 4 This is a comparative analysis diagram of the fault degree assessment error under different optimization methods in the embodiment of the present invention:

[0121] This figure compares the evaluation errors of various optimization methods under different fault severity levels. The horizontal axis represents fault severity (minor, moderate, and severe), and the vertical axis represents the relative error percentage of fault assessment (0-15%). The data shows that our proposed solution (diamonds) exhibits the lowest evaluation error across all fault severity levels: 2.9% for minor faults, 2.1% for moderate faults, and 1.3% for severe faults, with the average error remaining at 2.1%. In comparison, the errors of other methods are significantly higher: the error of the linear decreasing weight PSO (triangles) ranges from 3.8% to 5.8%, the error of the standard PSO (squares) ranges from 5.2% to 7.9%, and the parameter-free optimization method (circles) has the highest error, ranging from 7.2% to 11.5%. Our proposed solution reduces the error by 77.7% compared to the parameter-free optimization solution, 67.7% compared to the standard PSO, and 56.3% compared to the linear decreasing weight PSO. These data fully demonstrate that this technical solution has significant advantages in the accuracy of fault degree assessment, especially in dealing with severe faults, showing higher accuracy.

[0122] In an optional embodiment, a reducer remaining life prediction model is established, and the failure development trend of the reducer is predicted based on the reducer remaining life prediction model, and the generation of fault warning information includes:

[0123] Obtaining time points and status data from historical fault data of the reducer, establishing a reducer remaining life prediction model based on the time points and the status data, and optimizing the prediction model parameters by minimizing the sum of absolute errors between the predicted life value and the actual life value;

[0124] A state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and prediction confidence intervals are constructed by combining the future multi-step state prediction values ​​with process noise. The prediction confidence intervals are used to characterize the uncertainty range of the fault development trend;

[0125] Combining the predicted confidence interval with the current estimated remaining lifespan to construct a composite early warning indicator comprising a lifespan ratio term and a state change rate term, wherein the lifespan ratio term is the ratio of the current remaining lifespan to the lifespan threshold, and the state change rate term is the ratio of the state change speed to the speed threshold;

[0126] When the composite warning indicator is greater than a preset warning threshold, fault warning information is generated based on the current estimated remaining life and the fault development trend prediction result.

[0127] Obtain historical fault data for the reducer, including time points and status data. This historical fault data can be obtained from the reducer monitoring system. For example, three consecutive months of vibration signal data, temperature data, noise data, and oil analysis data can be obtained from the reducer of a manufacturing company. Specifically, the root mean square value of the reducer gearbox vibration can be selected as the status feature. Ten complete data sequences from normal operation to failure can be extracted from the historical data, with each data set containing status data at 200 time points.

[0128] Based on the acquired time points and state data, a prediction model for the remaining life of the reducer is established. In this embodiment, a double exponential trend model is used as the prediction model, which can be expressed as the relationship between the state value and time. In order to optimize the model parameters, the sum of the absolute errors between the predicted life value and the true life value is minimized in an iterative manner. Specifically, the initial parameter values ​​are set, for example, the attenuation coefficient in the model is initialized to 0.05, the growth coefficient is initialized to 0.02, and then the gradient descent method is used for parameter optimization. In actual applications, after training on 10 sets of historical data, the average prediction error of the model can be controlled within 7.5%, among which the attenuation coefficient is optimized to 0.038 and the growth coefficient is optimized to 0.025.

[0129] A state evolution equation is constructed based on the established prediction model. The state evolution equation describes how the device state evolves over time. The current state data of the reducer is input into the state evolution equation, and recursive iterative calculations are used to obtain the predicted values ​​for the future multi-step state. For example, for a certain operating reducer, its current root mean square vibration value is 2.8mm / s. The state evolution equation can predict the state change trend over the next 30 days. During the recursive calculation process, each predicted value is used as input for the next calculation. For example, if the predicted value on the first day is 2.85mm / s, it is used as input to predict the state value on the second day to be 2.91mm / s, and so on.

[0130] To characterize the uncertainty in the prediction process, the future multi-step state prediction values ​​are combined with process noise to construct a prediction confidence interval. Process noise can be determined by analyzing the volatility of historical data. In this embodiment, the standard deviation of the process noise is calculated by calculating the standard deviation of the historical data, which is 0.15 mm / s. The upper limit of the prediction confidence interval is the prediction value plus twice the standard deviation, and the lower limit is the prediction value minus twice the standard deviation. For example, if the prediction value on the 10th day is 3.25 mm / s, the prediction confidence interval is [2.95 mm / s, 3.55 mm / s]. This prediction confidence interval indicates that at a 95% confidence level, the actual state value of the reducer on the 10th day will fall within this interval.

[0131] The prediction confidence interval is combined with the current estimated remaining life to construct a composite early warning indicator. The composite early warning indicator includes a life ratio item and a state change rate item. The life ratio item is the ratio of the current remaining life to the life threshold, and the state change rate item is the ratio of the state change speed to the speed threshold. In this embodiment, based on the prediction model, the remaining life of the current reducer is estimated to be 25 days, and the life threshold is set to 30 days; the current state change speed is 0.06 mm / s / day, and the speed threshold is set to 0.05 mm / s / day. Therefore, the life ratio item is calculated as 25 / 30=0.833, and the state change rate item is calculated as 0.06 / 0.05=1.2. The composite early warning indicator can be obtained by weighted average calculation. In this example, the weights are set to 0.6 and 0.4 respectively, and the composite early warning indicator value is 0.833×0.6+1.2×0.4=0.98.

[0132] Set the warning threshold to 0.95. When the composite warning index value of 0.98 exceeds the warning threshold, the system generates a fault warning message. The warning message includes the following: "Device ID: GBX-1025, Current Status: Vibration value 2.8 mm / s exceeds the warning line, Estimated Remaining Life: 25 Days, Recommended Action: Planned Maintenance, Priority: Medium." The warning message also includes a status trend graph, visually displaying the status evolution trend and predicted confidence interval for the next 30 days.

[0133] Through the above method, the remaining life of the reducer can be accurately predicted, and fault warning information can be generated in a timely manner based on the prediction results, providing decision support for equipment maintenance, effectively avoiding unplanned equipment downtime, and improving equipment utilization and production efficiency.

[0134] Figure 5 This is a comparative analysis chart of fault identification accuracy under different working conditions in an embodiment of the present invention:

[0135] This figure compares the fault identification accuracy of three different methods under various operating conditions. The horizontal axis represents eight different operating conditions (low speed and light load, low speed and heavy load, medium speed and light load, medium speed and medium load, medium speed and heavy load, high speed and light load, high speed and medium load, and high speed and heavy load), while the vertical axis represents fault identification accuracy (70%-100%). The present technical solution (circles) maintains high identification accuracy across all operating conditions, ranging from 94% at low speed and light load to 87% at high speed and heavy load, demonstrating the most stable overall performance. The deep learning method (squares) performs second best, with accuracy fluctuating between 82% and 91%, but performance drops significantly to 82% under high speed and heavy load conditions. The traditional regression method (triangles) performs the worst, with overall low and highly fluctuating accuracy, gradually decreasing from 85% at low speed and light load to 71% at high speed and heavy load, with a particularly significant performance degradation under high speed conditions. Data shows that this technical solution maintains high recognition accuracy under various operating conditions, especially under complex high-speed and heavy-load conditions, demonstrating strong adaptability and reliability. This stable performance advantage has important practical significance for fault diagnosis of industrial equipment.

[0136] In an optional embodiment, a state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and the future multi-step state prediction values ​​are combined with process noise to construct a prediction confidence interval, including:

[0137] Establishing a state evolution equation based on the reducer remaining life prediction model, substituting a historical state sequence into the state evolution equation, calculating the square error between the predicted state value and the actual state value and summing them to obtain a square error sum, and determining the parameters of the state evolution equation by minimizing the square error sum;

[0138] Substituting the current state data and the current moment remaining life prediction value into the state evolution equation to obtain the next moment predicted state value, repeating the recursive iteration until a multi-step state prediction value of a preset number of steps is obtained;

[0139] Obtaining partial derivatives of state variables from the state evolution equation to obtain a Jacobian matrix, multiplying the Jacobian matrix by the multi-step state prediction value and accumulating the results to obtain a prediction error accumulation;

[0140] The prediction error accumulation is multiplied by the Jacobian matrix to obtain the prediction uncertainty, the square root of the prediction uncertainty is calculated and multiplied by the confidence level coefficient to obtain the confidence interval width, and the multi-step state prediction value is added or subtracted by the confidence interval width to obtain the prediction confidence interval.

[0141] Obtain historical operating data for the reducer, including condition monitoring data such as vibration signals, temperature data, and lubricant parameters. For example, for a planetary reducer in a mining facility, collect operating data for three consecutive months, sampling once an hour. This data includes parameters such as vibration RMS value, temperature, and input shaft speed. By preprocessing the raw data through denoising and normalization, a standardized historical state sequence is obtained, denoted as X_1, X_2, ..., X_n, where X_i represents the reducer state value at the i-th moment.

[0142] Based on the preprocessed historical state sequence, a model for predicting the remaining life of the reducer is established. This model can be constructed using machine learning methods such as BP neural networks, support vector regression, or random forests. Taking the BP neural network as an example, the input layer is set to 5 nodes, corresponding to 5 historical state values; the hidden layer is set to two layers, with 10 and 6 nodes respectively; and the output layer has 1 node, corresponding to the predicted remaining life value. By training this model, a mapping relationship between the current state and the remaining life is obtained.

[0143] Based on the aforementioned reducer remaining life prediction model, a state evolution equation is established. This equation describes the transition from the current state to the next state. This equation can be expressed as a function of the state value, its historical value, and the remaining life. In practice, the historical state values ​​X_t, X_(t-1), ..., X_(tp) and the current remaining life prediction value RUL_t are used as input variables, and X_(t+1) is used as the output variable to construct the general form of the state evolution equation.

[0144] To determine the specific parameters of the state evolution equation, the historical state sequence is substituted into the equation, and the squared errors between the predicted state values ​​and the actual state values ​​are calculated and summed. For example, for each set of adjacent state values ​​in the historical data, the state at the previous moment is substituted into the equation to obtain the predicted value, the square of the difference between the predicted value and the actual observed value is calculated, and then the total sum of squared errors is summed. By minimizing this sum of squared errors, the parameter values ​​in the state evolution equation can be determined. In actual calculations, optimization algorithms such as gradient descent or least squares method can be used to solve for the optimal parameters. Taking the data of a certain reducer as an example, the parameters of the state evolution equation obtained through training are: weight coefficients w_1=0.72, w_2=0.18, w_3=0.05, w_4=0.03, w_5=0.02, and a coefficient related to the remaining life of -0.015.

[0145] After determining the parameters of the state evolution equation, the state data X_t at the current time t and the current remaining life prediction value RUL_t are substituted into the state evolution equation to obtain the predicted state value X_(t+1) at the next time t+1. Then, X_(t+1) and the updated remaining life prediction value RUL_(t+1) are substituted into the state evolution equation again to obtain the predicted state value X_(t+2) at time t+2. And so on, through recursive iterative calculations, we ultimately obtain a multi-step state prediction value for a preset number of steps, such as a state prediction sequence for the next 24 hours.

[0146] To assess the uncertainty of the prediction results, it is necessary to construct a prediction confidence interval. First, the partial derivatives of the state evolution equation with respect to the state variables are calculated to obtain the Jacobian matrix. Specifically, the partial derivatives of the equation with respect to each input state variable are calculated to form the elements of the Jacobian matrix. For the above parameter example, the main elements of the Jacobian matrix are 0.72, 0.18, 0.05, 0.03, 0.02, and -0.015.

[0147] The Jacobian matrix is ​​multiplied by the multi-step state predictions and then accumulated to obtain the cumulative prediction error. This step reflects the propagation and accumulation of errors during the multi-step prediction process. In practice, for each prediction step, the Jacobian matrix is ​​multiplied by the current prediction error, and the results of all steps are accumulated to obtain the total cumulative prediction error.

[0148] Multiplying the cumulative forecast error by the Jacobian matrix yields the forecast uncertainty. This calculation accounts for the impact of model parameter uncertainty on the forecast. The square root of the resulting forecast uncertainty is then multiplied by the confidence level factor (for example, 1.96 for a 95% confidence level) to determine the confidence interval width. Finally, the confidence interval width is added or subtracted from the multi-step state forecast values ​​to form the upper and lower bounds of the forecast, known as the forecast confidence interval.

[0149] For example, let's take a 24-hour prediction for a reducer. The current state value is 0.65 (after normalization). Recursive calculation yields the following predicted state sequence for the next 24 hours: 0.67, 0.69, 0.72, 0.74, ..., 1.05. The corresponding 95% confidence interval widths are: 0.03, 0.05, 0.08, 0.11, ..., 0.32. Therefore, the 95% confidence interval for the state value after 24 hours is [0.73, 1.37].

[0150] Through the above method, multi-step prediction of the future state of the reducer can be achieved, and the confidence interval of the prediction results can be given, providing a reliable decision-making basis for preventive maintenance and fault warning of the equipment.

[0151] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0152] processor;

[0153] a memory for storing processor-executable instructions;

[0154] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0155] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0156] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting and diagnosing speed reducer faults in airport baggage systems based on digital twins, characterized by: include: Acquire real-time operating data of the reducer of the airport baggage system, including reducer speed data, vibration data, temperature data, current data, and noise data; Establishing a digital twin model of the airport baggage system reducer, and updating the digital twin model in real time using the real-time operating data; constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing features of the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model to perform fault type identification and fault severity assessment; Synchronizing the results of the fault type identification and the fault severity assessment with the digital twin model, dynamically optimizing the parameters of the digital twin model using a particle swarm optimization algorithm, establishing a reducer remaining life prediction model, predicting the fault development trend of the reducer based on the reducer remaining life prediction model, and generating fault warning information; Inputting the result of the fault type identification and the result of the fault severity assessment into the digital twin model, establishing a state mapping relationship between the result of the fault type identification, the result of the fault severity assessment and the digital twin model through a feature mapping function, and calculating a state synchronization error between the digital twin model and the actual system, wherein the state synchronization error is measured using a Frobenius norm; Constructing an optimization objective function based on the state mapping relationship and the state synchronization error, wherein the optimization objective function uses an adaptive weight coefficient to perform weighted fusion on the state mapping relationship and the state synchronization error, and the optimization objective function is used to optimize the parameter vector of the digital twin model; A particle swarm algorithm with dynamic inertia weight is used to calculate the dynamic neighborhood radius of the optimization objective function based on the number of iterations and the shrinkage factor. The particle fitness value is obtained by error fusion. The field influence factor is calculated according to the neighborhood center parameter and multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated by a linear combination of the local optimal position and the global optimal position. A maintenance plan is determined according to the fault warning information, the maintenance plan including maintenance time, maintenance content and maintenance priority, and the maintenance plan is sent to the airport baggage system control terminal.

2. The method according to claim 1, characterized in that Constructing a multidimensional feature vector based on the real-time operation data, extracting and fusing the multidimensional feature vector using a deep learning algorithm to obtain a fused feature vector, and inputting the fused feature vector into the digital twin model for fault type identification and fault severity assessment includes: Preprocessing the real-time operation data to obtain sampling data, and constructing a multidimensional feature vector including time domain features and frequency domain features based on the sampling data; Inputting the multidimensional feature vector into a multi-channel convolutional neural network, wherein each channel of the multi-channel convolutional neural network processes the time domain features and the frequency domain features respectively, extracting features from the multidimensional feature vector through a convolution layer and an activation function to obtain a feature map, and fusing the features of the feature map through an attention mechanism to obtain a fused feature vector, wherein the attention mechanism calculates an attention weight based on the query vector, the key vector, and the value vector; Based on the digital twin model, a deep neural network structure is adopted to identify the fault type of the fused feature vector through a softmax function and a ReLU activation function to obtain a result of the fault type identification; A health index is calculated based on a deviation between the fused feature vector and a pre-stored health status baseline feature. The health index is calculated by dividing the difference between the fused feature vector and the health status baseline feature by the health status baseline feature. The health index is input into a fuzzy inference system. The fuzzy inference system quantitatively evaluates the degree of fault based on a Gaussian membership function to obtain a fault degree evaluation result.

3. The method according to claim 2, characterized in that Based on the deep neural network structure adopted by the digital twin model, the fault type is identified on the fused feature vector through the softmax function and the ReLU activation function. The results of the fault type identification include: Input the fused feature vector into a deep neural network, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses a ReLU activation function for nonlinear mapping, and the output layer uses a softmax function to calculate the probability distribution of the fault type; The fault type is determined based on the probability distribution result, and the fault type is synchronized with the digital twin model. The fault type is verified and evaluated through the digital twin model to obtain the result of fault type identification.

4. The method according to claim 1, wherein The particle swarm algorithm with dynamic inertia weight is used to calculate the dynamic neighborhood radius of the optimization objective function based on the number of iterations and the shrinkage factor. The particle fitness value is obtained by error fusion. The field influence factor is calculated according to the neighborhood center parameter and multiplied by the basic inertia weight to obtain the dynamic inertia weight. The particle velocity is updated through a linear combination of the local optimal position and the global optimal position. The dynamic neighborhood radius is calculated based on the current number of iterations, the maximum number of iterations, and the shrinkage factor. The particle fitness value is calculated by combining the synchronization error, performance error, and regularization error of the digital twin model. The dynamic neighborhood radius and the particle fitness value are exponentially weighted and fused to obtain the adaptive neighborhood radius. Selecting digital twin model parameters that satisfy neighborhood constraints and have better optimization effects than the current particle as neighborhood individuals according to the adaptive neighborhood radius, and calculating neighborhood center parameters of the neighborhood individuals weighted by the inverse of the optimization effects. The neighborhood center parameters are used to guide the optimization direction of the digital twin model parameters; The neighborhood influence factor is calculated by the distance ratio between the neighborhood center parameter and the current digital twin model parameter and the global optimal parameter. The neighborhood influence factor is multiplied by the basic inertia weight to obtain the dynamic inertia weight of the fused neighborhood information. The basic inertia weight decreases nonlinearly with the iterative process. The particle velocity of the digital twin model parameters is updated based on the dynamic inertia weight of the fused neighborhood information.

5. The method according to claim 1, wherein Establishing a reducer remaining life prediction model, predicting the reducer failure development trend based on the reducer remaining life prediction model, and generating fault warning information including: Obtaining time points and status data from historical fault data of the reducer, establishing a reducer remaining life prediction model based on the time points and the status data, and optimizing the prediction model parameters by minimizing the sum of absolute errors between the predicted life value and the actual life value; A state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and prediction confidence intervals are constructed by combining the future multi-step state prediction values ​​with process noise. The prediction confidence intervals are used to characterize the uncertainty range of the fault development trend; Combining the predicted confidence interval with the current estimated remaining lifespan to construct a composite early warning indicator comprising a lifespan ratio term and a state change rate term, wherein the lifespan ratio term is the ratio of the current remaining lifespan to the lifespan threshold, and the state change rate term is the ratio of the state change speed to the speed threshold; When the composite warning indicator is greater than a preset warning threshold, fault warning information is generated based on the current estimated remaining life and the fault development trend prediction result.

6. The method according to claim 5, characterized in that A state evolution equation is constructed based on the reducer remaining life prediction model, current state data of the reducer is input into the state evolution equation, future multi-step state prediction values ​​are obtained through recursive iterative calculation, and the future multi-step state prediction values ​​are combined with process noise to construct a prediction confidence interval, including: Establishing a state evolution equation based on the reducer remaining life prediction model, substituting a historical state sequence into the state evolution equation, calculating the square error between the predicted state value and the actual state value and summing them to obtain a square error sum, and determining the parameters of the state evolution equation by minimizing the square error sum; Substituting the current state data and the current moment remaining life prediction value into the state evolution equation to obtain the next moment predicted state value, repeating the recursive iteration until a multi-step state prediction value of a preset number of steps is obtained; Obtaining partial derivatives of state variables from the state evolution equation to obtain a Jacobian matrix, multiplying the Jacobian matrix by the multi-step state prediction value and accumulating the results to obtain a prediction error accumulation; The prediction error accumulation is multiplied by the Jacobian matrix to obtain the prediction uncertainty, the square root of the prediction uncertainty is calculated and multiplied by the confidence level coefficient to obtain the confidence interval width, and the multi-step state prediction value is added or subtracted by the confidence interval width to obtain the prediction confidence interval.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images