Crane mislifting scene fault detection method, device and equipment and storage medium

Abstract

The application relates to a crane mis-hoisting scene fault detection method, device and equipment and a storage medium. The method comprises the following steps: collecting signal data of a crane in real time, calculating derived features of the crane according to the signal data, taking the signal data and the derived features as real feature sequences, generating a sample generation network based on the real feature sequences, obtaining generated fault samples, wherein the sample generation network is obtained based on a generative adversarial network, training a fault detection model based on the real feature sequences and the generated fault samples, and detecting the hoisting operation of the crane in real time based on the fault detection model. The application can make the fault diagnosis model effectively capture key abnormal signals such as power, rotating speed and lifting weight under different load working conditions, and improve the detection accuracy of the crane mis-hoisting fault.

Description

Technical Field

[0001] This application relates to the field of intelligent detection technology, and in particular to a fault detection method, device, equipment and storage medium for a crane mis-lifting scenario. Background Technology

[0002] Before unloading or loading operations, cranes in ports or docks need to manually disengage the locks between the cargo and the truck's carrying platform. If the locks are not fully disengaged due to operational errors, the crane may accidentally lift the truck along with the cargo, leading to equipment and cargo damage. Existing methods for detecting accidental lifting mostly focus on frequent faults such as rope fatigue, motor overload, and reducer malfunctions, with limited research on rare accidental lifting events such as accidentally lifting a truck. Because real-world rare accidental lifting events are scarce, the available fault samples are limited and the data distribution is uneven, resulting in low accuracy and a high risk of missed detection for existing automated detection schemes.

[0003] Therefore, how to improve the detection accuracy of rare mis-lifting faults of cranes during operation has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of the above, this application provides a fault detection method, device, equipment and storage medium for a crane mis-lifting scenario, the purpose of which is to solve the above-mentioned technical problems.

[0005] Firstly, this application provides a fault detection method for a crane mis-lifting scenario, the method comprising:

[0006] Real-time acquisition of crane signal data, and calculation of the crane's derived characteristics based on the signal data;

[0007] The signal data and the derived features are used as the real feature sequence. Based on the real feature sequence and the sample generation network, generated fault samples are obtained. The sample generation network is trained based on a generative adversarial network.

[0008] Based on the real feature sequence and the generated fault samples, a fault detection model is trained.

[0009] The crane's lifting operations are monitored in real time based on the fault detection model.

[0010] Secondly, this application provides a fault detection device for a crane mis-lifting scenario, the fault detection device comprising:

[0011] Acquisition module: Used to acquire signal data of the crane in real time and calculate the derived characteristics of the crane based on the signal data;

[0012] The generation module is used to take the signal data and the derived features as a real feature sequence, and obtain generated fault samples based on the real feature sequence and the sample generation network, wherein the sample generation network is trained based on a generative adversarial network.

[0013] Training module: used to train a fault detection model based on the real feature sequence and the generated fault samples;

[0014] Detection module: Used to detect the hoisting operation of the crane in real time based on the fault detection model.

[0015] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0016] Memory, used to store computer programs;

[0017] When the processor executes a program stored in the memory, it implements the steps of the fault detection method for a crane mis-lifting scenario as described in any embodiment of the first aspect.

[0018] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the fault detection method for a crane mis-lifting scenario as described in any embodiment of the first aspect.

[0019] The technical solutions provided in this application have the following advantages compared with the prior art:

[0020] This application acquires real-time signal data from the crane and calculates its derived features based on this data. Since these derived features more intuitively reflect the crane's operating status under different working conditions, they can better uncover potential information within the signal data, improving the accuracy and reliability of subsequent fault detection. By using the signal data and derived features as a true feature sequence, and generating fault samples based on this sequence and a sample generation network, the application avoids the problems of scarce real fault data and class imbalance in crane mis-lifting scenarios. A fault detection model is trained based on the true feature sequence and generated fault samples. This model is then used to detect crane lifting operations in real time, enabling the fault diagnosis model to effectively capture key abnormal signals such as power, speed, and lifting weight under different load conditions. This significantly improves the detection accuracy of rare mis-lifting faults during crane operation and keeps the false negative and false positive rates extremely low. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a preferred embodiment of the fault detection method for a crane mis-lifting scenario according to this application;

[0024] Figure 2 This is a schematic diagram of signal data in an embodiment of this application;

[0025] Figure 3 This is a schematic diagram of the TimeGAN network structure in an embodiment of this application;

[0026] Figure 4 A fault sample distribution map generated for embodiments of this application;

[0027] Figure 5 This is a schematic diagram of a preferred embodiment of the fault detection device for a crane mis-lifting scenario in this application;

[0028] Figure 6 This is a schematic diagram of a preferred embodiment of the electronic device of this application;

[0029] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0031] It should be noted that the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0032] Reference Figure 1 The diagram shown is a flowchart illustrating an embodiment of the fault detection method for a crane mis-lifting scenario according to this application. The method is executed by an electronic device, which can be implemented by a software system and / or a hardware system. The fault detection method for the crane mis-lifting scenario includes:

[0033] Step S10: Collect the crane's signal data in real time, and calculate the crane's derived characteristics based on the signal data;

[0034] Step S20: Using the signal data and the derived features as the real feature sequence, and based on the real feature sequence and the sample generation network, obtain the generated fault samples, wherein the sample generation network is trained based on a generative adversarial network;

[0035] Step S30: Based on the real feature sequence and the generated fault samples, train a fault detection model;

[0036] Step S40: Real-time detection of the crane's lifting operation based on the fault detection model.

[0037] In this embodiment, voltage and current sensors are pre-installed at the motor end of the crane to collect real-time signal data such as the three-phase voltage and current of the motor. The sampling frequency is no less than 2000 times per second. Hardware bandpass filtering is used to remove 50Hz interference and high-frequency noise from the power grid. Digital filtering can be performed via software, and the load level and operating status are recorded. Figure 2 The diagram shown is a schematic diagram of signal data in an embodiment of this application. The horizontal axis represents the number of sampling points, and the vertical axis represents the numerical value of the analog-to-digital conversion sampling.

[0038] Collect a large amount of normal lifting process data of the crane, simulate in a safe and controllable environment or obtain a small amount of data on truck malfunctions caused by accidental lifting from historical records, and mark the time period of the malfunction. This data can be used as part of the real feature sequence.

[0039] Since raw signal data (such as the three-phase voltage and current of the motor) can only reflect the basic operating state of the equipment and lacks sensitivity to the details of equipment operation and potential faults, it is difficult to capture subtle characteristic changes when a fault occurs if only raw signal data is relied upon for fault detection. Derived features, however, can more intuitively reflect the operating state of the crane under different working conditions. Therefore, by calculating derived features, the potential information in the data can be better extracted, thereby improving the accuracy and reliability of subsequent fault detection. Specifically, calculating the derived features of the crane based on the signal data includes:

[0040] The signal data is divided into several signal windows at preset time intervals;

[0041] Based on the crane's parameter information, the derived characteristics of the crane within each signal window are calculated, wherein the derived characteristics include: three-phase power, average power, motor torque, motor frequency, motor speed, motor slip, lifting height, and lifting weight.

[0042] The signal data can be segmented at preset time intervals (e.g., 3 seconds) to obtain several signal windows. Within each signal window, the derived characteristics of the lifting process are calculated, including three-phase power, average power, motor torque, motor frequency, motor speed, motor slip, lifting height, and lifting weight.

[0043] By analyzing the three-phase input voltage and current of the motor, the power factor can be calculated, thereby obtaining the motor's three-phase power and average power. With the motor model, number of pole pairs, and rated frequency determined, motor tests (such as no-load tests, locked-rotor tests, and DC resistance measurements) can be used to obtain stator resistance, stator leakage reactance, rotor equivalent resistance, rotor leakage reactance, main magnetic branch leakage reactance, and core losses. Once the motor parameters are fixed, the motor speed and slip can be directly calculated based on different operating conditions. After obtaining the motor's iron losses and mechanical losses through motor tests, the motor torque and frequency can be directly calculated based on different operating conditions.

[0044] After determining the motor speed, the lifting height of the goods can be calculated based on the initial height. With a fixed motor speed, the lifting weight is directly proportional to the motor power. Therefore, since the motor speed and power are directly proportional, the lifting weight can be calculated using linear fitting.

[0045] Because the probability of crane mis-lifting scenarios is low, real fault samples are scarce. This makes it difficult for fault detection models trained using traditional machine learning methods to obtain sufficient and effective fault samples for learning, thus affecting the performance of the detection models. Therefore, this embodiment uses a sample generation network based on generative adversarial networks (GANs) to expand the fault sample set. The collected real feature sequences (including signal data from normal samples, signal data from a small number of fault samples, and derived features) are input into the sample generation network. Through pre-training and fine-tuning, synthetic fault samples that are highly similar to real fault samples in statistical characteristics are generated. For example, assuming 100 real mis-lifting fault samples are obtained through simulation experiments, the sample generation network generates 1000 synthetic fault samples. These synthetic samples highly overlap with the real samples in terms of key feature distribution, effectively alleviating the problem of insufficient fault samples.

[0046] The sample generation network includes an embedder, a reconstructor, a generator, and a discriminator. The training process of the sample generation network includes:

[0047] The embedder and reconstructor are trained using a preset number of normal sample data so that the reconstructor can reconstruct the latent encoding generated by the embedder back to the original feature sequence.

[0048] A generator is trained using a random noise vector, a discriminator is trained using the feature sequences generated by the generator, and the parameters of the generator and discriminator are updated using a backpropagation algorithm.

[0049] During the training of the generator and discriminator, the parameters of the embedder and reconstructor are fixed. The parameters of the generator and discriminator are adjusted by using real fault samples combined with corresponding condition vectors, so that the generator generates samples that conform to the target distribution and the discriminator can distinguish between real samples and generated samples.

[0050] The sample generation network is trained based on the TimeGAN network, such as Figure 3 The diagram shown is a structural schematic of the TimeGAN network in an embodiment of this application, which mainly consists of four parts: an embedder, a reconstructor, a generator, and a discriminator.

[0051] In the TimeGAN network structure, the embedder maps the real feature sequence to the latent space to obtain the latent encoding. This is processed using a multi-layer gated recurrent unit (GRU), which learns the latent representation of the original data. The reconstructor reconstructs the latent representation back to the original feature space, ensuring that the embedding space can effectively reconstruct real samples. It employs a symmetric network structure. Therefore, the embedder and generator are structurally identical to the Variational Autoencoder (VAE). The generator takes a random noise vector and a static conditional vector (whether it is a faulty sample) as input and outputs a generated feature sequence. Its network consists of a combination of multiple Transformer layers and fully connected layers. The discriminator takes a real or generated sample as input and outputs a true / false discrimination probability. It is implemented using a temporal discriminant network (RNN + fully connected layers) to improve the ability to discriminate features of different lengths or details. Therefore, the generator and discriminator are structurally identical to the GAN network.

[0052] The loss function of the TimeGAN network is mainly divided into three parts: the VAE part consists of reconstruction loss, the GAN part consists of adversarial loss, and there is also a supervised loss part involving the temporal features of the data. During training, due to the relatively complex structure of the TimeGAN network, it is trained layer by layer in two steps: pre-training and fine-tuning. In the pre-training process, a large number of normal samples are used to train the embedder and reconstructor, enabling them to accurately reconstruct normal feature sequences. Simultaneously, the generator and discriminator are trained so that the generator can generate samples conforming to a normal distribution. In the fine-tuning process, the parameters of the embedder and reconstructor are fixed, and then a small number of real fault samples are used in conjunction with corresponding conditional vectors to fine-tune the generator, enabling it to generate fault samples under fault conditions. Simultaneously, the discriminator is trained to distinguish between real faults and generated fault samples.

[0053] Specifically, obtaining the generated fault samples based on the real feature sequence and the sample generation network includes:

[0054] The real feature sequence is input into the sample generation network to generate initial fault samples;

[0055] Calculate the similarity between the real fault samples in the true feature sequence and the initial fault samples;

[0056] From the initial fault samples, fault samples with similarity values ​​less than a preset threshold are removed to obtain candidate fault samples;

[0057] Based on preset physical verification rules, unreasonable samples are removed from the candidate fault samples to obtain the generated fault samples.

[0058] Since the fault samples generated by the sample generation network are likely unreasonable, they need to be screened. The fault samples generated by the network are designated as initial fault samples. An acceptable distance range is determined based on the distribution characteristics of real fault samples. The distributions of the initial fault samples and real fault samples are statistically analyzed across several key derived features, and the similarity of these distributions is calculated. A high similarity indicates that the generated initial fault samples fall within an acceptable distance range, while a low similarity indicates that they fall within an unacceptable distance range. Therefore, fault samples with similarity values ​​less than a preset threshold can be removed from the initial fault samples, thus eliminating fault samples outside the acceptable range and obtaining samples within the acceptable distance range (designated as candidate fault samples).

[0059] Next, physical rules are applied to the candidate fault samples to remove unreasonable samples, thus obtaining the generated fault samples. For example, within a fault sample window, the lifting height should remain monotonically constant or exhibit only reasonable minor fluctuations (small fluctuations caused by sensor noise are permissible); otherwise, it is considered a physically impossible sample. The motor speed and torque in the generated fault samples should not exceed the equipment design limits, and the trend of change should be smooth without sudden and unreasonable jumps. Power changes should be consistent with the load level, lifting speed, and other condition vectors to avoid generating samples that are significantly inconsistent with the given conditions. By calculating the distribution difference of key indicators between the generated fault samples and the actual fault samples, samples with large deviations can be eliminated. By applying preset physical verification rules (lifting height must be non-decreasing, motor speed cannot change abruptly), unreliable samples that do not conform to physical laws can be eliminated.

[0060] To verify the effectiveness and reliability of the generated fault samples in the feature space, this application embodiment also uses two dimensionality reduction visualization techniques, PCA (Principal Component Analysis) and t-SNE (t-Distributed Random Neighborhood Embedding), to evaluate the distribution of the generated fault samples and the real fault samples. Figure 4 The figure shows the distribution of fault samples generated in this application embodiment. The PCA evaluation results in the figure show that real fault samples are mainly concentrated in a certain region of the two-dimensional principal component space, while most of the generated fault samples are distributed within the real samples, exhibiting a high degree of overlap. This indicates that the generator has well reproduced the dominant feature distribution direction of the fault samples, demonstrating strong consistency at the global distribution level. The t-SNE evaluation results show that real fault samples form obvious feature clusters, while most of the generated fault samples are embedded within or at the boundaries of real sample clusters, indicating good similarity in their local feature relationships. A few generated fault samples show outliers, indicating room for improvement in extreme condition modeling. This demonstrates that the generated fault samples are highly consistent with real samples in their local structure, effectively supplementing the scarcity of real fault data.

[0061] A training set is constructed by mixing generated fault samples, real fault samples, and normal samples, while a separate validation set containing only real fault samples is retained. A fault detection model is then trained based on the training set. Specifically, training the fault detection model based on the real feature sequences and the generated fault samples includes:

[0062] The real feature sequence is mixed with the generated fault samples according to a preset ratio to obtain a mixed sample set;

[0063] The mixed sample set is converted into a training sample set in a three-channel format;

[0064] The EfficientNet network is trained based on the training sample set to obtain a fault detection model.

[0065] The real feature sequences and the generated fault samples are mixed at a preset ratio (e.g., 5:1) to obtain a mixed sample set. Since the original collected three-phase voltage, current, and other time-series data are in single-channel format, they are difficult to directly apply to deep convolutional networks. Therefore, this embodiment uses an image-based approach to expand the channels, converting the mixed sample set into a three-channel training sample set. The EfficientNet network is then trained based on the training sample set to obtain a fault detection model. In the design and training of the fault detection model, the EfficientNet network is used as the backbone feature extraction structure, and a multilayer perceptron (MLP) is integrated on top of it for subsequent classification tasks. The main function of the MLP structure is to perform further nonlinear mapping on the high-dimensional feature vectors, thereby obtaining a clearer fault category distribution during the training phase to optimize the model's discriminative ability.

[0066] By introducing generated fault data, the problem of imbalanced positive and negative samples is greatly alleviated. During the model training and evaluation phases, conventional performance metrics such as overall accuracy, AUC (Area Under Curve), false alarm rate, and missed detection rate can be used for evaluation to ensure that the model has high sensitivity and low false alarm rate in actual deployment.

[0067] By constructing the EfficientNet network structure as a fault detection model, it has efficient multi-scale feature extraction capabilities. Its main advantages include a composite scaling strategy for depth, width, and resolution, which can improve the model's ability to perceive multi-scale abnormal signals while keeping the model complexity under control. By constructing a representative and diverse training set through real fault samples and generated fault samples, the model's ability to identify rare abnormal samples can be improved.

[0068] Further, converting the mixed sample set into a three-channel format training sample set includes:

[0069] The time-domain signal data of the mixed samples is used as the original signal channel;

[0070] The frequency domain features of the time-domain signal data of the mixed samples are extracted by Fourier transform and used as the frequency domain channel;

[0071] The frequency domain features are subjected to logarithmic transformation and normalization to obtain logarithmic spectrum data, which serves as the spectrum enhancement channel.

[0072] The time-domain signal data of the mixed samples is used directly, preserving the original dynamic characteristics as the original signal channel. Frequency-domain features of the time-domain signal data of the mixed samples are extracted using Discrete Fourier Transform (DFT) as the frequency-domain channel, revealing hidden periodic features and anomalous frequency components. Logarithmic transformation and normalization are applied to the frequency-domain features to obtain log-DFT data, which serves as the spectrum enhancement channel. This enhances the discernibility of small-amplitude frequency components and further improves the balance of feature distribution.

[0073] Converting the mixed sample set into a three-channel format not only preserves the temporal characteristics of the signal but also introduces frequency domain features, enabling the model to more comprehensively understand the possible abnormal patterns in the signal. The complementary information between different channels helps deep learning models (such as EfficientNet) extract more discriminative features, thereby improving the accuracy and robustness of identifying mis-lifting faults.

[0074] After training, real-world examples of incorrect lifting that have not appeared in the historical data can be used as a validation set to evaluate the model's generalization ability under different loads, weather conditions, lifting frequencies, and other operating conditions. A comparative experiment was conducted using a container terminal crane as an example. The comparison results of the fault detection model on the same validation set are shown in the table below:

[0075]

[0076] As can be seen from the table, using the generated data significantly improves the model's accuracy and reduces the false negative rate, thus enhancing model performance.

[0077] To further improve the accuracy of the fault detection model, an initial alarm threshold was determined based on the fault probability distribution on the validation set and the results of small-scale field tests. This threshold is mainly divided into single-time thresholds and multiple-time confirmations. A single-time threshold triggers an alarm when the output probability of a window exceeds the threshold, while multiple-time confirmations automatically shut down the system when several consecutive windows exceed the threshold, thus reducing the risk of false alarms.

[0078] After the fault detection model is trained, the crane's lifting operation is detected in real time based on the fault detection model. The model can be exported to ONNX format and deployed to an industrial control computer. ONNX Runtime is used to accelerate inference on the target hardware, test the inference time, and ensure that the entire feature extraction and inference process meets the real-time requirements.

[0079] In one embodiment, the real-time detection of the crane's lifting operation based on the fault detection model includes:

[0080] The multilayer perceptron module in the fault detection model is removed to detect the crane's lifting operations in real time.

[0081] In practical deployment, to enhance the interpretability and controllability of the model in boundary decisions, while reducing model complexity and inference time, the MLP classification head used in the training phase was removed, and the feature vectors output by EfficientNet were directly used as the core basis for the fault detection model. This allows for manual control of the MLP threshold based on the characteristics of the dock / yard crane, increasing the model's generalization ability.

[0082] Reference Figure 5 The diagram shown is a functional module schematic of the fault detection device 100 for the crane mis-lifting scenario of this application.

[0083] The fault detection device 100 for the crane mis-lifting scenario described in this application is installed in an electronic device. Depending on its function, the fault detection device 100 includes a data acquisition module 110, a generation module 120, a training module 130, and a detection module 140. These modules can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and are stored in the memory of the electronic device.

[0084] In this embodiment, the functions of each module / unit are as follows:

[0085] Acquisition module 110: Used to acquire signal data of the crane in real time and calculate the derived characteristics of the crane based on the signal data;

[0086] Generation module 120: used to take the signal data and the derived features as real feature sequences, and obtain generated fault samples based on the real feature sequences and the sample generation network, wherein the sample generation network is trained based on a generative adversarial network;

[0087] Training module 130: used to train a fault detection model based on the real feature sequence and the generated fault samples;

[0088] Detection module 140: used to detect the hoisting operation of the crane in real time based on the fault detection model.

[0089] In one embodiment, calculating the derived characteristics of the crane based on the signal data includes:

[0090] The signal data of the crane is collected in real time, and the signal data is divided into several signal windows every preset time period;

[0091] Based on the crane's parameter information, the derived characteristics of the crane within each signal window are calculated, wherein the derived characteristics include: three-phase power, average power, motor torque, motor frequency, motor speed, motor slip, lifting height, and lifting weight.

[0092] In one embodiment, the sample generation network includes an embedder, a reconstructor, a generator, and a discriminator, and the training process of the sample generation network includes:

[0093] The embedder and reconstructor are trained using a preset number of normal sample data so that the reconstructor can reconstruct the latent encoding generated by the embedder back to the original feature sequence.

[0094] A generator is trained using a random noise vector, a discriminator is trained using the feature sequences generated by the generator, and the parameters of the generator and discriminator are updated using a backpropagation algorithm.

[0095] During the training of the generator and discriminator, the parameters of the embedder and reconstructor are fixed. The parameters of the generator and discriminator are adjusted by using real fault samples combined with corresponding condition vectors, so that the generator generates samples that conform to the target distribution and the discriminator can distinguish between real samples and generated samples.

[0096] In one embodiment, obtaining the generated fault samples based on the real feature sequence and the sample generation network includes:

[0097] The real feature sequence is input into the sample generation network to generate initial fault samples;

[0098] Calculate the similarity between the real fault samples in the true feature sequence and the initial fault samples;

[0099] From the initial fault samples, fault samples with similarity values ​​less than a preset threshold are removed to obtain candidate fault samples;

[0100] Based on preset physical verification rules, unreasonable samples are removed from the candidate fault samples to obtain the generated fault samples.

[0101] In one embodiment, training a fault detection model based on the real feature sequence and the generated fault samples includes:

[0102] The real feature sequence is mixed with the generated fault samples according to a preset ratio to obtain a mixed sample set;

[0103] The mixed sample set is converted into a training sample set in a three-channel format;

[0104] The EfficientNet network is trained based on the training sample set to obtain a fault detection model.

[0105] In one embodiment, converting the mixed sample set into a three-channel format training sample set includes:

[0106] The time-domain signal data of the mixed samples is used as the original signal channel;

[0107] The frequency domain features of the time-domain signal data of the mixed samples are extracted by Fourier transform and used as the frequency domain channel;

[0108] The frequency domain features are subjected to logarithmic transformation and normalization to obtain logarithmic spectrum data, which serves as the spectrum enhancement channel.

[0109] In one embodiment, the real-time detection of the crane's lifting operation based on the fault detection model includes:

[0110] The multilayer perceptron module in the fault detection model is removed to detect the crane's lifting operations in real time.

[0111] Reference Figure 6 The diagram shown is a schematic representation of a preferred embodiment of the electronic device of this application.

[0112] The electronic device includes a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0113] The memory 113 is used to store computer programs, such as a fault detection program for a crane mis-lifting scenario;

[0114] In some embodiments, the processor 111 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 111 is typically used to control the overall operation of the electronic device, such as performing data interaction or communication-related control and processing. In this embodiment, the processor 111 is used to run program code stored in the memory 113 or process data.

[0115] The communication interface 112 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The communication interface 112 may also be used to establish a communication connection between the electronic device and other electronic devices.

[0116] The memory 113 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 113 may be an internal storage unit of the electronic device, such as the hard disk or memory of the electronic device. In other embodiments, the memory 113 may also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. of the electronic device. Of course, the memory 113 may include both internal storage units and external storage devices of the electronic device. In this embodiment, the memory 113 is typically used to store the operating system and various computer programs installed on the electronic device, such as the program code of a fault detection program for a crane mis-lifting scenario. In addition, the memory 113 can also be used to temporarily store various types of data that have been output or will be output.

[0117] Figure 6 Only an electronic device with components 111-114 is shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0118] In one embodiment of this application, the processor 111, when executing the program stored in the memory 113, implements the fault detection method for a crane mis-lifting scenario provided in any of the foregoing method embodiments, including:

[0119] Real-time acquisition of crane signal data, and calculation of the crane's derived characteristics based on the signal data;

[0120] The signal data and the derived features are used as the real feature sequence. Based on the real feature sequence and the sample generation network, generated fault samples are obtained. The sample generation network is trained based on a generative adversarial network.

[0121] Based on the real feature sequence and the generated fault samples, a fault detection model is trained.

[0122] The crane's lifting operations are monitored in real time based on the fault detection model.

[0123] For a detailed explanation of the above steps, please refer to the above. Figure 1 This document describes a flowchart illustrating an embodiment of a fault detection method for a scenario where a crane accidentally lifts an object.

[0124] Furthermore, this application also proposes a computer-readable storage medium that is both non-volatile and volatile. This computer-readable storage medium is any one or any combination of several of the following: hard disk, multimedia card, SD card, flash memory card, SMC, read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disk read-only memory (CD-ROM), USB memory, etc. The computer-readable storage medium includes a data storage area and a program storage area. The program storage area stores a fault detection program for a crane mis-lifting scenario. When executed by a processor, the fault detection program for the crane mis-lifting scenario performs the following operations:

[0125] Real-time acquisition of crane signal data, and calculation of the crane's derived characteristics based on the signal data;

[0126] The signal data and the derived features are used as the real feature sequence. Based on the real feature sequence and the sample generation network, generated fault samples are obtained. The sample generation network is trained based on a generative adversarial network.

[0127] Based on the real feature sequence and the generated fault samples, a fault detection model is trained.

[0128] The crane's lifting operations are monitored in real time based on the fault detection model.

[0129] The specific implementation of the computer-readable storage medium in this application is largely the same as the specific implementation of the fault detection method for the above-mentioned crane mis-lifting scenario, and will not be described again here.

[0130] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware simulation platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0132] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A fault detection method for a crane mis-lifting scenario, characterized in that, The method includes: Real-time acquisition of crane signal data, and calculation of the crane's derived characteristics based on the signal data; The signal data and the derived features are used as the real feature sequence. Based on the real feature sequence and the sample generation network, generated fault samples are obtained. The sample generation network is trained based on a generative adversarial network. Based on the real feature sequence and the generated fault samples, a fault detection model is trained. The crane's lifting operations are monitored in real time based on the aforementioned fault detection model; The calculation of the crane's derived characteristics based on the signal data includes: The signal data of the crane is collected in real time, and the signal data is divided into several signal windows every preset time period; Based on the crane's parameter information, the derived characteristics of the crane within each signal window are calculated, wherein the derived characteristics include: three-phase power, average power, motor torque, motor frequency, motor speed, motor slip, lifting height, and lifting weight. The process of obtaining generated fault samples based on the real feature sequences and the sample generation network includes: The real feature sequence is input into the sample generation network to generate initial fault samples; Based on the distribution of real fault samples and initial fault samples in several key derived features in the real feature sequence, the similarity between real fault samples and initial fault samples in the real feature sequence is calculated. From the initial fault samples, fault samples with similarity values ​​less than a preset threshold are removed to obtain candidate fault samples; Based on preset physical verification rules, unreasonable samples are removed from the candidate fault samples to obtain the generated fault samples.

2. The fault detection method for a crane mis-lifting scenario as described in claim 1, characterized in that, The sample generation network includes an embedder, a reconstructor, a generator, and a discriminator. The training process of the sample generation network includes: The embedder and reconstructor are trained using a preset number of normal sample data so that the reconstructor can reconstruct the latent encoding generated by the embedder back to the original feature sequence. A generator is trained using a random noise vector, a discriminator is trained using the feature sequences generated by the generator, and the parameters of the generator and discriminator are updated using a backpropagation algorithm. During the training of the generator and discriminator, the parameters of the embedder and reconstructor are fixed. The parameters of the generator and discriminator are adjusted by using real fault samples combined with corresponding condition vectors, so that the generator generates samples that conform to the target distribution and the discriminator can distinguish between real samples and generated samples.

3. The fault detection method for a crane mis-lifting scenario as described in claim 1, characterized in that, The fault detection model is trained based on the real feature sequence and the generated fault samples, including: The real feature sequence is mixed with the generated fault samples according to a preset ratio to obtain a mixed sample set; The mixed sample set is converted into a training sample set in a three-channel format; The EfficientNet network is trained based on the training sample set to obtain a fault detection model.

4. The fault detection method for a crane mis-lifting scenario as described in claim 3, characterized in that, The step of converting the mixed sample set into a three-channel format training sample set includes: The time-domain signal data of the mixed samples is used as the original signal channel; The frequency domain features of the time-domain signal data of the mixed samples are extracted by Fourier transform and used as the frequency domain channel; The frequency domain features are subjected to logarithmic transformation and normalization to obtain logarithmic spectrum data, which serves as the spectrum enhancement channel.

5. The fault detection method for a crane mis-lifting scenario as described in any one of claims 1 to 4, characterized in that, The real-time detection of crane lifting operations based on the fault detection model includes: The multilayer perceptron module in the fault detection model is removed to detect the crane's lifting operations in real time.

6. A fault detection device for a crane mis-lifting scenario, characterized in that, The device includes: Acquisition module: Used to acquire signal data of the crane in real time and calculate the derived characteristics of the crane based on the signal data; The generation module is used to take the signal data and the derived features as a real feature sequence, and obtain generated fault samples based on the real feature sequence and the sample generation network, wherein the sample generation network is trained based on a generative adversarial network. Training module: used to train a fault detection model based on the real feature sequence and the generated fault samples; Detection module: used to detect the crane's lifting operations in real time based on the fault detection model; The calculation of the crane's derived characteristics based on the signal data includes: The signal data of the crane is collected in real time, and the signal data is divided into several signal windows every preset time period; Based on the crane's parameter information, the derived characteristics of the crane within each signal window are calculated, wherein the derived characteristics include: three-phase power, average power, motor torque, motor frequency, motor speed, motor slip, lifting height, and lifting weight. The process of obtaining generated fault samples based on the real feature sequences and the sample generation network includes: The real feature sequence is input into the sample generation network to generate initial fault samples; Based on the distribution of real fault samples and initial fault samples in several key derived features in the real feature sequence, the similarity between real fault samples and initial fault samples in the real feature sequence is calculated. From the initial fault samples, fault samples with similarity values ​​less than a preset threshold are removed to obtain candidate fault samples; Based on preset physical verification rules, unreasonable samples are removed from the candidate fault samples to obtain the generated fault samples.

7. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the fault detection method for a crane mis-lifting scenario as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault detection method for the crane mis-lifting scenario as described in any one of claims 1 to 5.

Images