Intelligent fault diagnosis and early warning system for electrical equipment

Abstract

The invention relates to the field of electrical equipment fault diagnosis, in particular to an intelligent fault diagnosis and early warning system for electrical equipment, which comprises a sensor data encoding module for generating a multi-time scale fusion feature from sensor data through a sensor encoder based on LSTM (Long Short Term Memory) and a self-attention mechanism architecture; the working condition data stress encoding module is used for generating working condition characteristics of the hoisting mechanism through an encoder according to the hoisting working condition data and the stress energy characteristics of the hoisting mechanism; the fault diagnosis module is used for generating a fault type of the electrical equipment through a fault diagnosis head based on an attention mechanism; the fault early warning module is used for carrying out fault trend early warning on the lifting electrical equipment through a detection head; and the vibration constraint module is used for performing training optimization based on the vibration constraint characteristics of the hoisting mechanism. According to the invention, it is ensured that the diagnosis and early warning result strictly follows the physical operation mechanism of the hoisting mechanism of the crane, and the fault diagnosis and early warning precision of the model under the working condition of frequent switching of hoisting electrical equipment of the hoisting mechanism is improved.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment fault diagnosis, and in particular to an intelligent fault diagnosis and early warning system for electrical equipment. Background Technology

[0002] In modern industrial production systems, crane hoisting mechanisms, as core equipment for material handling, are widely used in key areas such as ports, mines, and construction. Their operational stability directly determines production efficiency and operational safety. The hoisting electrical equipment of crane hoisting mechanisms, including the main hoisting motor, frequency converter, and brakes, serves as the power core of the hoist and operates under harsh environments of high load, strong vibration, and frequent switching of working conditions. During hoisting operations, drastic fluctuations in load weight, frequent adjustments to hoisting speed, and continuous impacts from mechanical vibrations can easily lead to problems such as insulation aging, contact wear, and brake failure in electrical equipment, causing production interruptions and economic losses. Therefore, efficient and accurate fault diagnosis and early warning for the hoisting electrical equipment of hoisting mechanisms has become a core requirement for the operation and maintenance of hoisting mechanisms.

[0003] With the increasing automation of hoisting mechanisms, some hoisting systems have introduced sensor-based monitoring systems for hoisting electrical equipment. These systems collect data from sensors such as current, voltage, and temperature to make preliminary fault diagnoses. However, at the sensor data processing level, traditional monitoring systems use feature extraction methods based on a single time scale, which cannot effectively integrate the dynamic characteristics of the equipment at different operating stages, resulting in insufficient feature representation capabilities. Furthermore, the operating conditions of hoisting mechanisms, including load changes and speed adjustments, are complex and variable. Traditional methods fail to deeply integrate operating condition data with sensor data, making fault diagnosis results susceptible to interference from operating conditions and significantly reducing accuracy.

[0004] Therefore, how to enable the lifting electrical equipment of the crane lifting mechanism to intelligently cope with the complex and ever-changing operating environment and achieve intelligent diagnosis and early warning of different faults is a technical problem that needs to be solved. Summary of the Invention

[0005] To address this, the present invention provides an intelligent fault diagnosis and early warning system for electrical equipment. By employing a physical information neural network (PINN) architecture that combines LSTM, self-attention mechanism, cluster encoder, fault diagnosis head, and convolutional dynamic threshold detection head, the vibration equation of the crane hoisting mechanism is embedded into the model as a hard constraint. This avoids the sensor data diagnosis results of the hoisting electrical equipment not conforming to physical laws due to sample bias caused by a purely data-driven model, ensuring that the diagnosis and early warning results strictly follow the physical operating mechanism of the hoisting mechanism, and improving the accuracy of fault diagnosis and early warning under frequent switching conditions of the hoisting electrical equipment.

[0006] To achieve the above objectives, this invention proposes an intelligent fault diagnosis and early warning system for electrical equipment, comprising: The sensor data encoding module is used to acquire sensor data of the lifting electrical equipment of the crane lifting mechanism, and to generate multi-timescale fusion features by passing the sensor data through a sensor encoder based on LSTM and self-attention mechanism architecture. The working condition data stress encoding module is used to obtain the lifting working condition data of the crane lifting mechanism, and to generate the lifting mechanism working condition characteristics by passing the lifting working condition data and the stress energy characteristics of the lifting mechanism through a cluster encoder. The fault diagnosis module is used to generate electrical equipment fault types by passing the multi-timescale fusion features, the hoisting mechanism operating condition features and vibration constraint features through a fault diagnosis head based on an attention mechanism and a convolutional output layer. The fault early warning module is used to combine the multi-timescale fusion features, the hoisting mechanism operating condition features, and the electrical equipment fault types through a convolutional dynamic threshold detection head to provide early warning of hoisting electrical equipment fault trends. The vibration constraint module is used to generate the vibration constraint features by passing the working condition data through the vibration equation of the hoisting mechanism, and to train and optimize the sensor encoder, the cluster encoder, the fault diagnosis head and the convolutional dynamic threshold detection head based on the vibration constraint features.

[0007] Furthermore, the sensor data encoding module includes: A multi-timescale feature extraction unit is used to process the sensor data through a multi-level LSTM layer to generate multi-timescale features; The attention time-scale fusion unit is used to generate time-scale attention weights by passing the multi-time-scale features through a self-attention mechanism, and to perform weighted fusion of the multi-time-scale features based on the time-scale attention weights to generate the multi-time-scale fused features. The sensor encoder includes the multi-level LSTM layer and the self-attention mechanism.

[0008] Furthermore, the multi-timescale feature extraction unit includes: A short-term feature extraction subunit is used to pass the sensor data through a first-level LSTM layer to generate instantaneous vibration load features; The intermediate feature extraction subunit is used to downsample the instantaneous vibration load features and the second LSTM layer to generate long-term velocity load regulation features; A long-term feature extraction subunit is used to downsample the long-term load regulation features of the speed and pass them through a third LSTM layer to generate equipment aging environment features; The multi-level LSTM layer includes a first-level LSTM layer, a second-level LSTM layer, and a third-level LSTM layer with sequentially increasing hidden neurons and sequentially increasing time windows. The multi-timescale features include the instantaneous vibration load features, the long-term speed load regulation features, and the equipment aging environment features.

[0009] Furthermore, the working condition data stress encoding module includes: The energy characteristic calculation unit is used to calculate stress parameters based on the hoisting mechanism load, cantilever torque, and motor torque, which are based on the stress energy characteristics of the hoisting mechanism. A clustering encoding unit is used to generate the hoisting mechanism's operating condition features by passing the spliced ​​vector of the hoisting condition data and the stress parameters through a clustering encoder based on convolutional linear transformation.

[0010] Furthermore, the energy characteristic calculation unit includes: The load stress calculation subunit is used to calculate load stress parameters based on the load of the hoisting mechanism and the set motor shaft cross-sectional area. A bending stress calculation subunit is used to calculate bending stress parameters based on the cantilever moment and cantilever section modulus. A torque stress calculation subunit is used to calculate torque stress parameters based on the motor torque and the section modulus of the motor output shaft. The stress parameters include the load stress parameters, the bending stress parameters, and the torque stress parameters.

[0011] In particular, by using a three-level LSTM layer to capture rapidly changing features such as vibration and instantaneous load, medium-speed features such as speed regulation and long-term load fluctuations, and slowly changing features such as equipment aging and environmental changes, the problem of misjudging operating condition fluctuations caused by traditional single-time-scale features is avoided. The self-attention mechanism dynamically weights the features of multiple time scales to ensure that when the system has faults at multiple time scales, it can accurately extract and fuse features. Based on the load of the hoisting mechanism, cantilever torque, and motor torque, load stress, bending stress, and torque stress parameters are calculated, and the stress parameters are encoded into the operating condition data through a clustering encoder. This can accurately identify hoisting mechanism operating conditions such as light load and slow speed, heavy load and fast speed, and no-load braking, avoiding misjudgment of faults under different operating conditions for the same parameter.

[0012] Furthermore, the fault diagnosis module includes: An attention fusion unit is used to combine the multi-timescale fusion features and the hoisting mechanism operating condition features through the attention mechanism of the fault diagnosis head to generate comprehensive fault features. The fault feature enhancement unit is used to perform gradient norm calculation on the multi-timescale fused features to generate multi-timescale fault enhancement features. The fault diagnosis unit is used to generate the electrical equipment fault type by passing the concatenated vector of the comprehensive fault features, the multi-timescale fault enhancement features and the vibration constraint features through a convolutional output layer.

[0013] Furthermore, the fault early warning module includes: The dynamic threshold calculation unit is used to calculate the fault threshold adjustment degree according to the fault type of the electrical equipment, and to generate the working condition threshold adjustment degree by passing the working condition characteristics of the hoisting mechanism through the convolutional linear mapping layer of the dynamic threshold detection head. The dynamic threshold is calculated based on the fault threshold adjustment degree and the working condition threshold adjustment degree. The fault warning judgment unit is used to determine whether to issue a fault trend warning for the hoisting electrical equipment based on a comparison between the fault type of the electrical equipment and the dynamic threshold.

[0014] Furthermore, the vibration constraint module includes: The vibration stiffness constraint unit is used to generate vibration stiffness constraint characteristics by passing the vibration stiffness sub-equation of the vibration equation of the hoisting mechanism through the mass of the hoisting electrical equipment, the vibration data of the sensor data, the damping coefficient, the dynamic stiffness coefficient and the foundation excitation force. The vibration energy constraint unit is used to generate vibration energy constraint characteristics by passing the stress parameters, the damping coefficient, the vibration data, and the basic excitation force through the vibration energy sub-equation of the hoisting mechanism vibration equation; The vibration constraint feature calculation unit is used to perform a weighted summation of the vibration stiffness constraint feature and the vibration energy constraint feature to generate the vibration constraint feature; The operating data includes the mass of the lifting electrical equipment and the basic excitation force.

[0015] Furthermore, the vibration stiffness constraint unit includes: The stiffness coefficient calculation subunit is used to perform a linear transformation on the operating characteristics of the hoisting mechanism to generate the dynamic stiffness coefficient.

[0016] Furthermore, the vibration energy constraint unit is used to subtract the product of the derivative of the vibration data and the basic excitation force from the derivative of the stress parameter, and then add the product of the second derivative of the vibration data and the damping coefficient to generate the vibration energy constraint feature.

[0017] In particular, by using the attention mechanism of the fault diagnosis head, dynamic weight allocation is performed on the multi-timescale fusion features and the hoisting mechanism operating condition features to avoid fault misjudgment caused by operating condition confusion. The gradient norm calculation of the multi-timescale fusion features can amplify the feature differences of early weak faults, so as to achieve accurate fault judgment and early warning of electrical equipment such as motor overload, insulation aging and frequency converter faults.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention adopts a physical information neural network PINN architecture with LSTM, self-attention mechanism, cluster encoder, fault diagnosis head and convolutional dynamic threshold detection head. The vibration equation of the crane hoisting mechanism is used as a hard constraint embedded model. This avoids the sensor data diagnosis results of the hoisting electrical equipment that do not conform to physical laws due to sample bias in the pure data driven model. It ensures that the diagnosis and early warning results strictly follow the physical operation mechanism of the hoisting mechanism and improves the fault diagnosis and early warning accuracy of the model under the frequent switching conditions of the hoisting electrical equipment.

[0019] In particular, this invention captures rapidly changing features such as vibration and instantaneous load, medium-speed features such as speed regulation and long-term load fluctuation, and slowly changing features such as equipment aging and environmental changes through a three-level LSTM layer. This avoids the problem of misjudging operating condition fluctuations caused by traditional single-time-scale features. By dynamically weighting multi-time-scale features through a self-attention mechanism, it ensures that when the system experiences multi-time-scale faults, it can accurately extract and fuse features. Based on the hoisting mechanism load, cantilever torque, and motor torque, it calculates load stress, bending stress, and torque stress parameters. The stress parameters are then encoded into the operating condition data through a clustering encoder, which can accurately identify hoisting mechanism operating conditions such as light load slow speed, heavy load fast speed, and no-load braking, avoiding misjudgment of faults under different operating conditions for the same parameter.

[0020] In particular, this invention uses the attention mechanism of the fault diagnosis head to dynamically assign weights to the multi-timescale fusion features and the hoisting mechanism operating condition features, avoiding misjudgment of faults caused by confusion of operating conditions. The gradient norm calculation of the multi-timescale fusion features can amplify the feature differences of early weak faults, enabling accurate fault judgment and early warning of electrical equipment such as motor overload, insulation aging, and frequency converter faults. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of an intelligent fault diagnosis and early warning system for electrical equipment according to an embodiment of the present invention; Figure 2 This is a schematic flowchart of a sensor encoder for an intelligent fault diagnosis and early warning system for electrical equipment according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the cluster encoder of an intelligent fault diagnosis and early warning system for electrical equipment according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the process for generating vibration constraint features in an intelligent fault diagnosis and early warning system for electrical equipment according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0025] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] like Figures 1 to 4 As shown, this invention provides an intelligent fault diagnosis and early warning system for electrical equipment. By employing a physical information neural network (PINN) architecture that combines LSTM, self-attention mechanism, cluster encoder, fault diagnosis head, and convolutional dynamic threshold detection head, the vibration equation of the crane hoisting mechanism is embedded as a hard constraint in the model. This avoids the sensor data diagnosis results of the hoisting electrical equipment that do not conform to physical laws due to sample bias in a purely data-driven model, ensuring that the diagnosis and early warning results strictly follow the physical operating mechanism of the hoisting mechanism, and improving the fault diagnosis and early warning accuracy of the model under the frequent switching conditions of the hoisting electrical equipment.

[0027] like Figure 1 As shown, the intelligent fault diagnosis and early warning system for electrical equipment according to an embodiment of the present invention includes: The sensor data encoding module is used to acquire sensor data of the lifting electrical equipment of the crane lifting mechanism, and to generate multi-timescale fusion features by passing the sensor data through a sensor encoder based on LSTM and self-attention mechanism architecture. The working condition data stress encoding module is used to obtain the lifting working condition data of the crane lifting mechanism, and to generate the lifting mechanism working condition characteristics by passing the lifting working condition data and the stress energy characteristics of the lifting mechanism through a cluster encoder. The fault diagnosis module is used to generate electrical equipment fault types by passing the multi-timescale fusion features, the hoisting mechanism operating condition features and vibration constraint features through a fault diagnosis head based on an attention mechanism and a convolutional output layer. The fault early warning module is used to combine the multi-timescale fusion features, the hoisting mechanism operating condition features, and the electrical equipment fault types through a convolutional dynamic threshold detection head to provide early warning of hoisting electrical equipment fault trends. The vibration constraint module is used to generate the vibration constraint features by passing the working condition data through the vibration equation of the hoisting mechanism, and to train and optimize the sensor encoder, the cluster encoder, the fault diagnosis head and the convolutional dynamic threshold detection head based on the vibration constraint features.

[0028] In particular, the aforementioned sensor encoder, cluster encoder, fault diagnosis head, and convolutional dynamic threshold detection head together constitute an intelligent fault diagnosis and early warning model for electrical equipment based on the Physical Information Neural Network (PINN) architecture. The process of generating multi-timescale fusion features, hoisting mechanism operating condition features, electrical equipment fault types, and early warning of hoisting electrical equipment fault trends is the forward propagation process of this model, while the generation of the vibration constraint features is the backward propagation optimization process of this model.

[0029] like Figure 2 As shown, in this embodiment, the sensor data encoding module includes: A multi-timescale feature extraction unit is used to process the sensor data through a multi-level LSTM layer to generate multi-timescale features; The attention time-scale fusion unit is used to generate time-scale attention weights by passing the multi-time-scale features through a self-attention mechanism, and to perform weighted fusion of the multi-time-scale features based on the time-scale attention weights to generate the multi-time-scale fused features. The sensor encoder includes the multi-level LSTM layer and the self-attention mechanism.

[0030] Specifically, the process of generating multi-timescale fused features through the self-attention mechanism can be represented as follows: In the formula, , , These represent the attention weights at multiple learnable time scales, and Sigmoid represents the Sigmoid function. , , These represent the weight parameter matrices for learnable self-attention mechanisms adapted to different time scales. A concatenated vector representing features across multiple time scales. This indicates element-wise multiplication. This represents the fusion features across multiple time scales.

[0031] In particular, by using different numerical distributions of multiple weight parameter matrices, the sensitivity to instantaneous features is enhanced, the correlation between instantaneous anomalies and medium-term trends is adapted, and the long-term state is reflected. This allows the proportion of the three types of self-attention weights in the multi-timescale fusion features to be adjusted, making the multi-timescale fusion features suitable for different feature timescales such as rapid identification of urgent faults, judgment of medium-term faults, and prediction of long-term fault trends. This enables full utilization of a small number of fault samples of hoisting electrical equipment in hoisting mechanisms.

[0032] like Figure 2 As shown, in this embodiment, the multi-timescale feature extraction unit includes: A short-term feature extraction subunit is used to pass the sensor data through a first-level LSTM layer to generate instantaneous vibration load features; The intermediate feature extraction subunit is used to downsample the instantaneous vibration load features and the second LSTM layer to generate long-term velocity load regulation features; A long-term feature extraction subunit is used to downsample the long-term load regulation features of the speed and pass them through a third LSTM layer to generate equipment aging environment features; The multi-level LSTM layer includes a first-level LSTM layer, a second-level LSTM layer, and a third-level LSTM layer with sequentially increasing hidden neurons and sequentially increasing time windows. The multi-timescale features include the instantaneous vibration load features, the long-term speed load regulation features, and the equipment aging environment features.

[0033] In particular, by using three LSTM layers to capture fast-changing features such as vibration and instantaneous load, medium-speed features such as speed regulation and long-term load fluctuation, and slow-changing features such as equipment aging and environmental changes, the problem of misjudging operating condition fluctuations by traditional single-time-scale features is avoided. By using a self-attention mechanism to dynamically weight the features of multiple time scales, the system can accurately extract and fuse features when multiple time-scale faults occur.

[0034] Specifically, the process of generating multi-timescale features using multi-level LSTM layers can be represented as: In the formula, , , These represent the instantaneous vibration load characteristics, long-term speed load regulation characteristics, and equipment aging environment characteristics, respectively. , , These represent the first, second, and third LSTM layers, respectively, with their corresponding time windows preferably being 1s, 60s, and 3600s. Since they respectively need to possess the capabilities of short-term rapid response, medium-term trend capture, and long-term periodic trend capture, their corresponding hidden neurons are preferably 64, 128, and 256. Represents sensor data, , These represent the downsampling characteristics of instantaneous vibration load characteristics and the downsampling characteristics of long-term velocity load regulation characteristics, respectively.

[0035] Specifically, the sensor data includes at least the instantaneous values ​​of the three-phase input current of the motor of the hoisting electrical equipment of the hoisting mechanism, the instantaneous values ​​of the three-phase input voltage of the motor, the output frequency of the frequency converter, the DC bus voltage of the frequency converter, the switching frequency of the frequency converter, the operating efficiency of the frequency converter, and the temperature of the motor bearing.

[0036] Therefore, the limitation of a single LSTM being unable to adapt to multi-frequency sensor data is avoided by using multi-level LSTM layers.

[0037] like Figure 3 As shown, in this embodiment, the working condition data stress encoding module includes: The energy characteristic calculation unit is used to calculate stress parameters based on the hoisting mechanism load, cantilever torque, and motor torque, which are based on the stress energy characteristics of the hoisting mechanism. A clustering encoding unit is used to pass the spliced ​​vector of the working condition data and the stress parameters through a clustering encoder based on convolutional linear transformation to generate the working condition features of the hoisting mechanism.

[0038] Please continue reading. Figure 3 As shown, in this embodiment, the energy characteristic calculation unit includes: The load stress calculation subunit is used to calculate load stress parameters based on the load of the hoisting mechanism and the set motor shaft cross-sectional area. A bending stress calculation subunit is used to calculate bending stress parameters based on the cantilever moment and cantilever section modulus. A torque stress calculation subunit is used to calculate torque stress parameters based on the motor torque and the section modulus of the motor output shaft. The stress parameters include the load stress parameters, the bending stress parameters, and the torque stress parameters.

[0039] In particular, the load stress, bending stress, and torque stress parameters are calculated based on the lifting mechanism load, cantilever torque, and motor torque. The stress parameters are then encoded into the working condition data through a cluster encoder, which can accurately identify the lifting mechanism working conditions such as light load slow speed, heavy load fast speed, and no-load braking, avoiding misjudgment of faults under different working conditions with the same parameter.

[0040] Specifically, the process by which the working condition data stress encoding module generates the working condition characteristics of the hoisting mechanism can be represented as follows: In the formula, Indicates the load stress parameter, , , These represent the load stress parameters, bending stress parameters, and torsional stress parameters, respectively. , These represent the cantilever moment (in N or kN) and the cantilever section modulus (in m²), respectively. , These represent the motor torque (in N·m or kN·m) and the motor output shaft section modulus (in m³), ​​respectively. This indicates the motor torque (unit: N·m or kN·m). This indicates the cross-sectional modulus of the motor output shaft (in m³). This indicates the operating characteristics of the hoisting mechanism. express Activation function , These represent the cluster encoder's data on operating conditions. and stress parameters spliced ​​vector Learnable weight matrix and bias terms for performing convolutional linear transformation.

[0041] Specifically, the operating data includes the load weight of the hoisting mechanism, load rate (the ratio of load weight to rated load), load duration, standard deviation of load weight fluctuation, vertical movement speed of the hook, and lateral movement speed of the trolley.

[0042] In this embodiment, the fault diagnosis module includes: An attention fusion unit is used to combine the multi-timescale fusion features and the hoisting mechanism operating condition features through the attention mechanism of the fault diagnosis head to generate comprehensive fault features. The fault feature enhancement unit is used to perform gradient norm calculation on the multi-timescale fused features to generate multi-timescale fault enhancement features. The fault diagnosis unit is used to generate the electrical equipment fault type by passing the concatenated vector of the comprehensive fault features, the multi-timescale fault enhancement features and the vibration constraint features through a convolutional output layer.

[0043] In particular, by using the attention mechanism of the fault diagnosis head, dynamic weight allocation is performed on the multi-timescale fusion features and the hoisting mechanism operating condition features to avoid fault misjudgment caused by operating condition confusion. The gradient norm calculation of the multi-timescale fusion features can amplify the feature differences of early weak faults, so as to achieve accurate fault judgment and early warning of electrical equipment such as motor overload, insulation aging and frequency converter faults.

[0044] Specifically, the process by which the fault diagnosis head generates electrical equipment fault types can be represented as follows: In the formula, , , These represent the query mapping vector, key mapping vector, and value mapping vector of the attention mechanism, respectively. , These represent the learnable weight matrix and bias term of the mapped query vector, respectively. , Let represent the learnable weight matrix and bias term of the mapped key vector, respectively. , Let represent the learnable weight matrix and bias term of the mapped value vector, respectively. This represents the multi-timescale fusion features. This indicates the operating characteristics of the hoisting mechanism. Indicates the comprehensive fault characteristics, express function, This indicates the dimension value of the query mapping vector. The gradient norm value represents the multi-timescale fusion feature, that is, the norm value of the first-order partial derivative of each element in the direction of the instantaneous vibration load feature, the direction of the long-term velocity load regulation feature, and the direction of the equipment aging environment feature. Indicates the type of electrical equipment fault. express Activation function , These represent the learnable convolutional weight matrix and bias term of the convolutional output layer, respectively.

[0045] In this embodiment, the fault early warning module includes: The dynamic threshold calculation unit is used to calculate the fault threshold adjustment degree according to the fault type of the electrical equipment, and to generate the working condition threshold adjustment degree by passing the working condition characteristics of the hoisting mechanism through the convolutional linear mapping layer of the dynamic threshold detection head. The dynamic threshold is calculated based on the fault threshold adjustment degree and the working condition threshold adjustment degree. The fault warning judgment unit is used to determine whether to issue a fault trend warning for the hoisting electrical equipment based on a comparison between the fault type of the electrical equipment and the dynamic threshold.

[0046] Specifically, the process of generating a dynamic threshold can be represented as: In the formula, , These represent a combination of operating condition threshold adjustment and fault threshold adjustment. This represents the learnable weight matrix of the convolutional linear mapping layer. This indicates the operating characteristics of the hoisting mechanism. This represents the adjustment factor, preferably 0.32. This represents the norm value of the first-order partial derivative of the multi-timescale fused feature in the time direction. This represents the basic threshold, preferably 0.3. This indicates a dynamic threshold.

[0047] like Figure 4 As shown, in this embodiment, the vibration constraint module includes: The vibration stiffness constraint unit is used to generate vibration stiffness constraint characteristics by passing the vibration stiffness sub-equation of the vibration equation of the hoisting mechanism through the mass of the hoisting electrical equipment, the vibration data of the sensor data, the damping coefficient, the dynamic stiffness coefficient and the foundation excitation force. The vibration energy constraint unit is used to generate vibration energy constraint characteristics by passing the stress parameters, the damping coefficient, the vibration data, and the basic excitation force through the vibration energy sub-equation of the hoisting mechanism vibration equation; The vibration constraint feature calculation unit is used to perform a weighted summation of the vibration stiffness constraint feature and the vibration energy constraint feature to generate the vibration constraint feature; The operating data includes the mass of the lifting electrical equipment and the basic excitation force.

[0048] like Figure 4 As shown, in this embodiment, the vibration stiffness constraint unit includes: The stiffness coefficient calculation subunit is used to perform a linear transformation on the operating characteristics of the hoisting mechanism to generate the dynamic stiffness coefficient.

[0049] Specifically, the vibration stiffness sub-equation can be expressed as: In the formula, Indicates the characteristics of vibration stiffness constraint. The mass of the lifting system of the hoisting mechanism is preferably the sum of the masses of the motor, transmission components, and load. This represents the foundation stiffness coefficient, preferably 0.4. Representing vibration data, This represents the damping coefficient, which is determined based on the structure of the lifting system. Indicates the dynamic stiffness coefficient. Indicates the material's inherent stiffness coefficient. This represents the basic excitation force, which is preferably the sum of the load force and electromagnetic force of the lifting electrical equipment.

[0050] In particular, by adapting the vibration stiffness constraint characteristics to the dynamic changes in working conditions, when the lifting condition is a heavy-load lifting, the load stress parameters in the working condition characteristics of the lifting mechanism increase and the dynamic stiffness coefficient increases, so that the vibration stiffness constraint characteristics can accurately match the characteristics of the enhanced stiffness of the equipment under heavy load, and filter the interference of heavy-load vibration on sensor data.

[0051] like Figure 4 As shown, in this embodiment, the vibration energy constraint unit is used to subtract the product of the derivative of the vibration data and the basic excitation force from the derivative of the stress parameter, and then add the product of the second derivative of the vibration data and the damping coefficient to generate the vibration energy constraint feature.

[0052] Specifically, the vibrational energy quantum equation can be expressed as: In the formula, Indicates the characteristics of vibration energy constraint. Indicates stress parameters, This represents the basic excitation force, preferably the sum of the load force and electromagnetic force of the lifting electrical equipment. Representing vibration data, , Let represent the derivative of the vibration data and the second derivative of the vibration data, respectively, that is, the rate of change and the acceleration of the vibration data. This represents the damping coefficient, which is determined based on the structure of the lifting system.

[0053] In particular, vibration energy constraint characteristics quantify the correlation of fault vibrations. For example, when the vibration energy increases due to the wear of motor bearings, the derivative of the stress parameter increases, and the vibration energy constraint characteristics increase accordingly. This characteristic can then be used to determine whether the vibration anomaly is caused by an electrical fault, rather than simply mechanical vibration.

[0054] In this embodiment, by employing a physical information neural network (PINN) architecture that combines LSTM, self-attention mechanism, cluster encoder, fault diagnosis head, and convolutional dynamic threshold detection head, the vibration equation of the crane hoisting mechanism is embedded as a hard constraint model. This avoids the pure data-driven model from producing sensor data diagnostic results for the hoisting electrical equipment that do not conform to physical laws due to sample bias. It ensures that the diagnostic and early warning results strictly follow the physical operating mechanism of the hoisting mechanism, thereby improving the accuracy of fault diagnosis and early warning under the frequent switching conditions of the hoisting electrical equipment. By employing a three-level LSTM layer to capture rapidly changing features such as vibration and instantaneous load, medium-speed features such as speed regulation and long-term load fluctuations, and slowly changing features such as equipment aging and environmental changes, the system avoids the problem of misjudging operating condition fluctuations caused by traditional single-time-scale features. A self-attention mechanism dynamically weights multi-time-scale features to ensure accurate extraction and fusion of features when multi-time-scale faults occur. Based on the hoisting mechanism load, cantilever torque, and motor torque, load stress, bending stress, and torque stress parameters are calculated. These stress parameters are then encoded into the operating condition data using a clustering encoder, accurately identifying hoisting mechanism operating conditions such as light load slow speed, heavy load fast speed, and no-load braking, avoiding misjudgments of faults under different operating conditions for the same parameter. Through the attention mechanism of the fault diagnosis head, dynamic weight allocation is applied to the multi-time-scale fused features and hoisting mechanism operating condition features to avoid misjudgments caused by operating condition confusion. Gradient norm calculation of the multi-time-scale fused features amplifies the feature differences of early, subtle faults, enabling accurate fault diagnosis and early warning for electrical equipment such as motor overload, insulation aging, and inverter faults.

[0055] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0056] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent fault diagnosis and warning system for electrical equipment, characterized in that, The method comprises the following steps: a sensor data encoding module is used to obtain sensor data of hoisting electrical equipment of a hoisting mechanism of a crane, and the sensor data is input into a sensor encoder based on an LSTM and a self-attention mechanism architecture to generate multi-time scale fusion features; a working condition data stress encoding module is used to obtain hoisting working condition data of the hoisting mechanism of the crane, and the hoisting working condition data and hoisting mechanism stress energy features are input into a clustering encoder to generate hoisting mechanism working condition features; a fault diagnosis module is used to input the multi-time scale fusion features, the hoisting mechanism working condition features and vibration constraint features into a fault diagnosis head based on an attention mechanism and a convolution output layer to generate electrical equipment fault types; a fault early warning module is used to input the multi-time scale fusion features, the hoisting mechanism working condition features and the electrical equipment fault types into a convolution dynamic threshold detection head to perform hoisting electrical equipment fault trend early warning; a vibration constraint module is used to input the working condition data into a hoisting mechanism vibration equation to generate vibration constraint features, and the sensor encoder, the clustering encoder, the fault diagnosis head and the convolution dynamic threshold detection head are trained and optimized based on the vibration constraint features.

2. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 1 wherein, The sensor data encoding module comprises: a multi-time scale feature extraction unit is used to input the sensor data into a multi-level LSTM layer to generate multi-time scale features; an attention time scale fusion unit is used to input the multi-time scale features into a self-attention mechanism to generate time scale attention weights, and the multi-time scale features are weighted and fused based on the time scale attention weights to generate the multi-time scale fusion features; wherein the sensor encoder comprises the multi-level LSTM layer and the self-attention mechanism.

3. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 2 wherein, The multi-time scale feature extraction unit comprises: a short-term feature extraction subunit is used to input the sensor data into a first-level LSTM layer to generate vibration instantaneous load features; a medium-term feature extraction subunit is used to input the vibration instantaneous load features into a downsampling and a second LSTM layer to generate speed long-term load adjustment features; a long-term feature extraction subunit is used to input the speed long-term load adjustment features into a downsampling and a third LSTM layer to generate device aging environment features; wherein the multi-level LSTM layer comprises the first-level LSTM layer, the second LSTM layer and the third LSTM layer with increasing hidden neurons and time windows, and the multi-time scale features comprise the vibration instantaneous load features, the speed long-term load adjustment features and the device aging environment features.

4. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 1 wherein, The working condition data stress encoding module comprises: an energy feature calculation unit is used to calculate stress parameters based on hoisting mechanism load, cantilever torque and motor torque of hoisting mechanism stress energy features; a clustering encoding unit is used to input a splicing vector of the hoisting working condition data and the stress parameters into a clustering encoder based on convolution linear transformation to generate the hoisting mechanism working condition features.

5. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 4 wherein, The energy feature calculation unit comprises: The load stress calculation subunit is configured to calculate a load stress parameter based on the lifting mechanism load and a set motor shaft sectional area; The bending stress calculation subunit is configured to calculate a bending stress parameter based on the cantilever torque and a cantilever sectional modulus; The torque stress calculation subunit is configured to calculate a torque stress parameter based on the motor torque and a motor output shaft sectional modulus; The stress parameters include the load stress parameter, the bending stress parameter, and the torque stress parameter.

6. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 1 wherein, The fault diagnosis module includes: The attention fusion unit is configured to fuse the multi-time scale fusion feature and the lifting mechanism working condition feature through an attention mechanism of the fault diagnosis head to generate a comprehensive fault feature; The fault feature enhancement unit is configured to perform gradient norm calculation on the multi-time scale fusion feature to generate a multi-time scale fault enhancement feature; The fault diagnosis unit is configured to pass the comprehensive fault feature, the multi-time scale fault enhancement feature, and a splicing vector of the vibration constraint feature through a convolution output layer to generate the electrical equipment fault type.

7. The intelligent fault diagnosis and warning system for electrical equipment according to any one of claims 1 to 6, characterized in that, The fault early warning module includes: The dynamic threshold calculation unit is configured to calculate a fault threshold adjustment degree according to the electrical equipment fault type, pass the lifting mechanism working condition feature through a convolution linear mapping layer of a convolution dynamic threshold detection head to generate a working condition threshold adjustment degree, and calculate a dynamic threshold based on the fault threshold adjustment degree and the working condition threshold adjustment degree; The fault early warning judgment unit is configured to determine whether to perform lifting electrical equipment fault trend early warning based on a comparison of the electrical equipment fault type and the dynamic threshold.

8. The intelligent fault diagnosis and warning system for electrical apparatus as claimed in claim 4 wherein, The vibration constraint module includes: The vibration stiffness constraint unit is configured to pass the lifting electrical equipment mass, vibration data of the sensor data, a damping coefficient, a dynamic stiffness coefficient, and a base excitation force through a vibration stiffness sub-equation of the lifting mechanism vibration equation to generate a vibration stiffness constraint feature; The vibration energy constraint unit is configured to pass the stress parameter, the damping coefficient, the vibration data, and the base excitation force through a vibration energy sub-equation of the lifting mechanism vibration equation to generate a vibration energy constraint feature; The vibration constraint feature calculation unit is configured to perform weighted summation on the vibration stiffness constraint feature and the vibration energy constraint feature to generate the vibration constraint feature; The working condition data includes the lifting electrical equipment mass and the base excitation force.

9. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 8 wherein, The vibration stiffness constraint unit includes: The stiffness coefficient calculation subunit is configured to perform linear transformation on the lifting mechanism working condition feature to generate the dynamic stiffness coefficient.

10. The intelligent fault diagnosis and warning system for electrical equipment as claimed in claim 8 wherein, The vibration energy constraint unit is configured to subtract a product of a derivative of the stress parameter and the vibration data from a derivative of the vibration data and the base excitation force, and add a product of a second derivative of the vibration data and the damping coefficient to generate the vibration energy constraint feature.

Images