Equipment fault automatic detection and early warning method, device, equipment and medium

Abstract

The invention relates to the field of fault diagnosis, and discloses an equipment fault automatic detection and early warning method and device, equipment and a medium, and the method comprises the steps: obtaining current fault monitoring data and historical fault monitoring data for target port equipment; inputting the current fault monitoring data and the historical fault monitoring data into the trained backhaul type coding and decoding network to obtain a fault detection result; performing early warning when the fault detection result shows that the target port equipment is in an abnormal operation state; the backhaul type coding and decoding network comprises an input coding branch, a historical coding branch and a decoding branch, the input coding branch is used for carrying out multistage feature coding on current fault monitoring data, and the historical coding branch is used for generating historical feature embedding vectors corresponding to historical fault monitoring data; and the decoding branch is used for performing multi-level feature decoding after splicing the output of the input coding branch and the historical feature embedded vector features to obtain a fault detection result. According to the application, the problems of automatic monitoring and fault early warning of the port equipment are solved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of fault diagnosis, and in particular to a method, apparatus, equipment and medium for automatic detection and early warning of equipment faults. Background Technology

[0002] The construction of smart ports is a significant trend in current port development. It fully leverages the deep integration of next-generation information technology with port construction to achieve optimized resource allocation and efficient, convenient operation. However, as ports become increasingly intelligent, control systems are becoming more complex. The 24 / 7 on-call maintenance approach is no longer sufficient for automated operations, leading to increased workload for maintenance personnel and increased operational risks and instability. Therefore, there is an urgent need to achieve automatic monitoring and fault early warning for intelligent equipment to save labor costs, improve operational efficiency, reduce risks, and support the sustainable development of smart ports. Summary of the Invention

[0003] The purpose of this invention is to provide at least one method, device, equipment and medium for automatic detection and early warning of equipment faults, which can at least solve the problem of automatic monitoring and fault early warning of port equipment, and at least achieve the effects of saving labor costs, improving operation and maintenance efficiency and reducing risks.

[0004] To address the aforementioned technical problems, at least one embodiment of this application provides an automatic equipment fault detection and early warning method, comprising: acquiring current fault monitoring data and historical fault monitoring data for a target port equipment; inputting the current fault monitoring data and historical fault monitoring data into a trained backhaul codec network to obtain a fault detection result; and issuing an early warning when the fault detection result indicates that the target port equipment is in an abnormal operating state; wherein, the backhaul codec network includes an input encoding branch, a historical encoding branch, and a decoding branch, the input encoding branch being used to perform multi-level feature encoding on the current fault monitoring data, the historical encoding branch being used to generate historical feature embedding vectors corresponding to historical fault monitoring data, and the decoding branch being used to concatenate the output of the input encoding branch and the historical feature embedding vectors and then perform multi-level feature decoding to obtain the fault detection result.

[0005] At least one embodiment of this application also provides an automatic equipment fault detection and early warning device, comprising: a data acquisition module for acquiring current fault monitoring data and historical fault monitoring data for a target port equipment; a fault detection module for inputting the current fault monitoring data and historical fault monitoring data into a trained backhaul codec network to obtain a fault detection result; and an anomaly early warning module for issuing an early warning when the fault detection result indicates that the target port equipment is in an abnormal operating state; wherein, the backhaul codec network includes an input encoding branch, a historical encoding branch, and a decoding branch, the input encoding branch for performing multi-level feature encoding on the current fault monitoring data, the historical encoding branch for generating historical feature embedding vectors corresponding to historical fault monitoring data, and the decoding branch for performing multi-level feature decoding after concatenating the output of the input encoding branch and the historical feature embedding vectors to obtain the fault detection result.

[0006] At least one embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described automatic device fault detection and early warning method.

[0007] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described automatic detection and early warning method for equipment faults.

[0008] The automatic equipment fault detection and early warning method provided in this application utilizes a backhaul codec network to simultaneously analyze equipment status based on current fault monitoring data and recent historical fault monitoring data, achieving comprehensive and reliable automatic fault detection and early warning for port equipment. Backhaul codec is achieved through a continuous codec structure and the reuse of historical data. By simultaneously considering historical damage monitoring data and current instantaneous status data in the port equipment operating status analysis, the codec network can capture the changing trends of port equipment status over time, helping to discover potential faults or anomalies that are difficult to detect using only real-time data. Furthermore, combining historical fault monitoring data allows for a more comprehensive assessment of the equipment's health status. The continuous codec structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection.

[0009] In some optional embodiments, the input encoding branch includes at least three encoder levels to perform multi-level feature encoding on the current fault monitoring data, and the decoding branch includes at least three decoders to perform multi-level feature decoding by concatenating the output of the input encoding branch and historical feature embedding vector features. Furthermore, the feature encoding results output by each encoder level are fused during the multi-level feature decoding process. This continuous encoder-decoder structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection. Fusing the feature encoding results output by each encoder level during the multi-level feature decoding process further enhances the accuracy of fault detection.

[0010] In some optional embodiments, the input encoding branch includes a three-level encoder, and the decoding branch includes a three-level decoder. The input encoding branch includes a first embedding module, a first encoder, a second encoder, and a third encoder; the historical encoding branch includes a second embedding module; and the decoding branch includes a first splicing module, a first decoder, a second splicing module, a second decoder, a third splicing module, a third decoder, a fourth splicing module, a linear layer, and a classification layer. The first embedding module is used to convert the current fault monitoring data into the current feature embedding vector and then input it into the first encoder. After multi-level feature encoding by the first encoder, the second encoder and the third encoder, the output is generated. The second embedding module is used to convert historical fault monitoring data into historical feature embedding vectors and then output them. The first concatenation module is used to concatenate the output of the input encoding branch and the historical feature embedding vector; The first decoder is used to decode the splicing result of the first splicing module. The second splicing module is used to splice the decoding result of the first decoder with the feature encoding result of the third encoder. The second decoder is used to decode the splicing result of the second splicing module. The third splicing module is used to splice the decoding result of the second decoder with the feature encoding result of the second encoder. The third decoder is used to decode the splicing result of the third splicing module. The fourth splicing module is used to splice the decoding result of the third decoder with the feature encoding result of the first encoder. The linear layer is used to perform a linear transformation on the splicing result of the fourth splicing module. The classification layer is used to output the fault detection result based on the linear transformation result.

[0011] In some optional embodiments, the automatic equipment fault detection and early warning method further includes: conducting a fault monitoring data simulation and acquisition experiment of the target port equipment, debugging different states of the target port equipment, recording fault monitoring data before and after the fault occurs in the fault state, and fault monitoring data in the non-fault state, determining the fault monitoring data before and after the fault occurs as a fault data group, determining the fault monitoring data in the non-fault state as a normal data group, and the fault data group and the normal data group constitute a fault detection dataset; training the backhaul encoding and decoding network based on the fault detection dataset.

[0012] In some optional embodiments, the fault detection result includes a first score and a second score, wherein when the first score exceeds a preset value, it indicates that the target port equipment is in an abnormal operating state.

[0013] In some optional embodiments, the encoder and the decoder adopt the same structure, including: Multi-head attention layer, used to extract multi-dimensional attention features from the input; The first normalization layer is used to normalize the sum of the output and input of the multi-head attention layer. The feedforward layer is used to perform a linear transformation on the normalized result. It uses a nonlinear function to activate the linear transformation result before performing the linear transformation. The second normalization layer is used to normalize the sum of the output and input of the feedforward layer.

[0014] In some optional embodiments, the first embedding module and the second embedding module generate word embedding vectors using the Word2vec word vector model, calculate a position encoding vector for each word embedding vector, and add the word embedding vector and the position encoding vector element by element to obtain the corresponding feature embedding vector. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0016] Figure 1 This is a flowchart of an automatic equipment fault detection and early warning method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the codec structure provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the return-to-home codec network provided in an embodiment of this application; Figure 4 This is a schematic diagram of the backhaul codec network construction process provided in an embodiment of this application; Figure 5This is a schematic diagram of an automatic equipment fault detection and early warning device provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0018] To address the aforementioned technical problems of automatic monitoring and fault early warning for port equipment, this invention proposes an automatic equipment fault detection and early warning method. The implementation details of the automatic equipment fault detection and early warning method in this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.

[0019] Example 1: The automatic fault detection and early warning method for equipment in this embodiment can be applied to electronic devices with communication, computing, and data storage capabilities. Its specific process can be as follows: Figure 1 As shown, it includes: Step 101: Obtain current damage monitoring data and historical damage monitoring data for the target port equipment.

[0020] Specifically, the target port equipment includes dockside ship loaders, yard reclaimers, stacker-reclaimers, and stacker cranes, etc. Damage monitoring data includes time-series information on damage monitoring for various types of port equipment. Time-series information refers to sequential data arranged according to the order of occurrence. Current damage monitoring data refers to the damage monitoring data at the current moment, while historical damage monitoring data includes at least the damage monitoring data from the previous moment.

[0021] In some examples, historical damage monitoring data includes damage monitoring data from n time points prior to the current time, where n ≥ 1, and the previous n time points are consecutive. Taking the scenario of structural damage monitoring for various types of port equipment as an example, fault monitoring data refers to the monitoring data obtained by monitoring structural damage to the target port equipment. The technologies used in structural damage monitoring include one of acoustic emission technology, ultrasonic guided wave technology, and strain response technology. The corresponding monitoring data are acoustic emission monitoring data, ultrasonic guided wave monitoring data, and strain monitoring data. These data are all time-series information, that is, sequential data arranged according to the order of occurrence in time. In practical applications, other technologies can also be used for structural damage monitoring of port equipment, not limited to the examples above.

[0022] In practical applications, this method is used to perform acoustic emission damage monitoring tasks (various defects may generate acoustic emission, including crack formation, plastic deformation, and phase transformation, etc. By detecting and analyzing the high-frequency sound waves emitted by the material when it is deformed or stressed through acoustic emission testing, corresponding fault monitoring data can be obtained). Before the structural damage fault occurs, that is, when the equipment is in normal condition, the acoustic emission signal detected by the acoustic emission sensor is a low-frequency noise signal, that is, it only contains background noise information; after the structural damage fault occurs, that is, when the equipment is in a fault state, the acoustic emission sensor detects structural damage, and at this time the acoustic emission signal contains the structural damage signal.

[0023] Step 102: Input the current damage monitoring data and historical damage monitoring data into the trained backhaul encoder-decoder network to obtain the fault detection results.

[0024] Specifically, the backhaul coding-decoding network includes an input coding branch, a history coding branch, and a decoding branch. The input coding branch is used to perform multi-level feature encoding on the current damage monitoring data. The history coding branch is used to generate historical feature embedding vectors corresponding to historical damage monitoring data. The decoding branch is used to concatenate the output of the input coding branch and the historical feature embedding vectors and then perform multi-level feature decoding to obtain the fault detection result.

[0025] Step 103: When the fault detection results indicate that the target port equipment is in an abnormal operating state, an early warning is issued.

[0026] This embodiment utilizes a backhaul codec network to simultaneously analyze the port equipment status based on current damage monitoring data and recent historical fault monitoring data, achieving comprehensive and reliable automatic fault detection and early warning for port equipment. Backhaul codec is achieved through a continuous codec structure and the reuse of historical fault monitoring data. By simultaneously considering historical and current damage monitoring data in the port equipment operating status analysis, the codec network can capture the changing trends of the port equipment status over time, helping to discover potential faults or anomalies that are difficult to detect using only real-time fault monitoring data. Furthermore, combining historical fault monitoring data allows for a more comprehensive assessment of the equipment's health status. The continuous codec structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection.

[0027] In some embodiments, the input encoding branch includes at least three encoder levels to perform multi-level feature encoding on the current damage monitoring data, and the decoding branch includes at least three decoders to perform multi-level feature decoding by concatenating the output of the input encoding branch and historical feature embedding vector features. Furthermore, the feature encoding results output by each encoder level are fused during the multi-level feature decoding process. This continuous encoder-decoder structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection. Fusing the feature encoding results output by each encoder level during the multi-level feature decoding process further enhances the accuracy of fault detection.

[0028] In some embodiments, the input encoding branch includes a three-level encoder, and the decoding branch includes a three-level decoder. The input encoding branch includes a first embedding module, a first encoder, a second encoder, and a third encoder; the historical encoding branch includes a second embedding module; and the decoding branch includes a first concatenation module, a first decoder, a second concatenation module, a second decoder, a third concatenation module, a third decoder, a fourth concatenation module, a linear layer, and a classification layer. The first embedding module is used to convert the current damage monitoring data into the current feature embedding vector and then input it into the first encoder. After multi-level feature encoding by the first encoder, the second encoder and the third encoder, the output is generated. The second embedding module is used to convert historical damage monitoring data into historical feature embedding vectors and then output them. The first concatenation module is used to concatenate the output of the input encoding branch and the historical feature embedding vector; The first decoder is used to decode the splicing result of the first splicing module. The second splicing module is used to splice the decoding result of the first decoder with the feature encoding result of the third encoder. The second decoder is used to decode the splicing result of the second splicing module. The third splicing module is used to splice the decoding result of the second decoder with the feature encoding result of the second encoder. The third decoder is used to decode the splicing result of the third splicing module. The fourth splicing module is used to splice the decoding result of the third decoder with the feature encoding result of the first encoder. The linear layer is used to perform a linear transformation on the splicing result of the fourth splicing module. The classification layer is used to output the fault detection result based on the linear transformation result.

[0029] In the actual implementation, all encoders and decoders use the same structure, and all feature splicing modules have the same internal structure.

[0030] In some embodiments, the first embedding module and the second embedding module generate word embedding vectors using the Word2vec word vector model, calculate a position encoding vector for each word embedding vector, and add the word embedding vector and the position encoding vector element by element to obtain the corresponding feature embedding vector. The Word2vec word vector model converts words in natural language into low-dimensional dense vectors that can be processed by computers, and allows the vectors to capture the semantic and grammatical relationships between words. For fault monitoring data of target port equipment, it is essentially the information expressed by words in natural language, such as the acoustic emission signal at time i. Using the Word2vec word vector model to convert the input acoustic emission signal at time i into a word embedding vector is a specific implementation of converting words in natural language into low-dimensional dense vectors that can be processed by computers. A word embedding vector is a type of low-dimensional dense vector that can be processed by computers.

[0031] The function of the first or second embedded module is to process the fault monitoring data output by the port equipment into the input format of the backhaul codec network in real time, while ensuring that the model can identify the fault information in the fault monitoring data of the port equipment.

[0032] In some examples, taking the first embedding module as an example, the current fault monitoring data is first converted into corresponding word embedding vectors. This step directly uses the Word2vec word vector model to generate word embedding vectors for the fault monitoring data. Then, a position encoding vector related to the position of each word embedding vector is calculated. The position encoding vector of each word embedding vector is calculated according to the following formula. Finally, the generated position encoding vector is added element by element to the input word embedding vector, thereby incorporating position information into the input and obtaining the final network input vector.

[0033] In the positional encoding vector calculation formula, the word position is p, the word embedding dimension is z, the model embedding dimension is d, and the even-dimensional positional embedding PE(p, 2z) and the odd-dimensional positional embedding PE(p, 2z+1) are calculated as follows:

[0034]

[0035] Here, the word position p refers to the position of the element in the current fault monitoring data sequence, and the word embedding dimension z and the embedding dimension d are both fixed values.

[0036] In some embodiments, the encoder and decoder include: Multi-head attention layer, used to extract multi-dimensional attention features from the input; The first normalization layer is used to normalize the sum of the output and input of the multi-head attention layer. The feedforward layer is used to perform a linear transformation on the normalized result. It uses a nonlinear function to activate the linear transformation result before performing the linear transformation. The second normalization layer is used to normalize the sum of the output and input of the feedforward layer.

[0037] In some examples, the encoder and decoder use the same structure, such as Figure 2 As shown. The input S of the codec is the output feature of the previous module, with a dimension equal to the model embedding dimension d. The codec construction process is as follows: First, the input S is fed into a multi-head attention layer for multi-dimensional attention feature extraction to fully extract feature information; then, the output of the multi-head attention layer is added to the input S itself, and the result is normalized by a normalization layer to obtain a normalized value D; next, the normalized value D is fed into a feedforward layer, which first performs a linear transformation on it, then activates it using the ReLU nonlinear function, and then performs another linear transformation. This step is to map the input normalized value D to a higher-dimensional space, then filters it using the ReLU nonlinear function, and finally transforms it back to the original dimension; the output of the previous step is added to the normalized value D itself, and the result is normalized by a normalization layer to obtain the final output F of the codec. Since the feature dimension is adjusted in the feedforward layer, the output F has the same dimension as the input S, which is d.

[0038] In some examples, the overall structure of a backhaul codec network is as follows: Figure 3 As shown, it mainly includes the following three steps: input encoding branch design, historical encoding branch design, and decoding branch design.

[0039] The input encoding branch structure is as follows: The fault monitoring data at the current moment is input into the first embedding module for preprocessing to generate the network input vector; the network input vector of dimension d is sequentially input into the first, second, and third encoders that are connected end to end to obtain the encoded outputs B1, B2, and B3 of dimension d, respectively. The historical encoding branch structure is as follows: historical fault monitoring data is input into the second embedding module for preprocessing, thereby generating historical feature embedding vectors; The decoding branch structure is as follows: The encoded output B3 of the third encoder and the historical feature embedding vector are input into the first feature concatenation module for feature concatenation. According to the embedding module, the embedding dimension is d. Therefore, in the first feature concatenation module, the encoded output B3 of the third encoder and the historical feature embedding vector are added together by dimension, and then normalized to output feature Q1 with dimension d.

[0040] Feature Q1 is input into the first decoder to obtain decoded feature J1; decoded feature J1 and encoded feature B3 from the third encoder are then input into the second feature concatenation module to obtain feature Q2 of dimension d; feature Q2 is input into the second decoder to obtain decoded feature J2; decoded feature J2 and encoded feature B2 from the second encoder are then input into the third feature concatenation module to obtain feature Q3 of dimension d; feature Q3 is input into the third decoder to obtain decoded feature J3; decoded feature J3 and encoded feature B1 from the first encoder are then input into the fourth feature concatenation module to obtain feature Q4 of dimension d.

[0041] Finally, the d-dimensional feature Q4 is output to the linear layer and then subjected to a fully connected layer linear transformation, resulting in an output dimension of 2, which is then output to the classification layer. In this embodiment, the classification layer uses the softmax function, and the output is 2-dimensional data, representing the probabilities of the input data being normal and abnormal data, respectively.

[0042] The above describes the construction process of a return-to-work encoder-decoder network. The input to the encoder and decoder is the output of the previous module, and the output of each encoder and decoder is its encoded result.

[0043] In some embodiments, the above-described automatic equipment fault detection and early warning method further includes: Step 201: Conduct a simulation and acquisition experiment of fault monitoring data for the target port equipment, debug different states of the target port equipment, record the damage monitoring data before and after the fault occurs in the fault state, and the damage monitoring data in the non-fault state. The damage monitoring data before and after the fault occurs is determined as the fault data group, and the damage monitoring data in the non-fault state is determined as the normal data group. The fault data group and the normal data group constitute the fault detection dataset.

[0044] Specifically, fault monitoring data simulation experiments are conducted to determine different fault states. Fault monitoring data before the fault occurs is recorded as historical fault monitoring data, and fault monitoring data after the fault occurs is recorded as current fault monitoring data. The pairs of historical and current fault monitoring data are labeled as fault data groups. In addition, a large number of historical and current fault monitoring data pairs under normal conditions of the port equipment are recorded and labeled as normal data groups. After obtaining a sufficient number of fault and normal data groups, a fault detection dataset is constructed.

[0045] In practice, erosion from the climate and long-term operational wear can cause corrosion, cracks, and plastic deformation in port equipment, leading to performance degradation. Fault signal acquisition methods are divided into two types: 1. Normal data acquisition on real, intact equipment, and fault data acquisition on existing, damaged equipment; 2. Fabricating a large number of simulated structural components, artificially creating corrosion, cracks, and plastic deformation on these components to achieve fault monitoring data acquisition before and after a fault occurs. In this embodiment, a fault monitoring data simulation acquisition experiment is conducted, artificially creating corrosion, cracks, and plastic deformation on simulated structural components to achieve various fault states such as corrosion, cracks, and plastic deformation, thereby enabling fault monitoring data acquisition before and after a fault occurs.

[0046] In some cases, "sufficient quantity" can refer to a sample size of 10,000 or more, enough to support backpropagation encoding / decoding network training until the loss function converges.

[0047] Step 202: Train a backhaul encoder-decoder network based on the fault detection dataset.

[0048] In some examples, the backhaul codec network is trained on the fault detection dataset. During the training process, the ratio of fault data group to normal data group is set to 4:1 to ensure that the backhaul codec network pays more attention to the characteristics of fault data. After the training converges, the final backhaul codec network is obtained.

[0049] In practice, the construction process of a backhaul codec network is as follows: Figure 4 As shown, firstly, a fault detection dataset is constructed; then, a device signal embedding module, a codec, and a backhaul codec network are designed, and finally, a backhaul codec network is generated.

[0050] In some embodiments, the fault detection result includes a first score and a second score. When the first score exceeds a preset value, it indicates that the target port equipment is in an abnormal operating state.

[0051] In the specific implementation, the first score and the second score are the abnormal data score and the normal data score, respectively. In practical applications, the abnormal data score and the normal data score can represent the probability of a fault occurring and the probability of no fault occurring, respectively. A backhaul codec model is used to monitor the equipment status in real time. The damage monitoring data of the port equipment is input into the backhaul codec network in real time, and the corresponding abnormal data score and normal data score are output. This allows us to obtain the probability of the target port equipment malfunctioning and the probability of no fault occurring. When the abnormal data score is greater than a preset value (for example, it is set to 0.5 based on experience, that is, the probability of a fault exceeding 50%, then an alarm is triggered; in practical applications, this preset value can also be manually adjusted according to the characteristics of the dataset and experimental results), an alarm is triggered, indicating that the current port equipment may be in an abnormal operating state.

[0052] This embodiment of the method for automatic equipment fault detection and early warning based on a backhaul codec network analyzes equipment status simultaneously using current fault monitoring data and recent historical fault monitoring data. This enables automatic monitoring and fault early warning of intelligent equipment, saving labor costs, improving operational efficiency, reducing risks, and supporting the sustainable development of smart port construction. The beneficial effects of this method are: (1) Compared with the 24 / 7 human-managed operation and maintenance method, the backhaul codec network can monitor the equipment status in real time without human intervention and provide real-time early warning of potential risks in equipment operation. It can significantly save manpower costs, improve operation and maintenance efficiency, and effectively ensure the safe and efficient operation of port equipment.

[0053] (2) Compared to ordinary fault detection networks, which only use a single current instantaneous state output for analysis, the backhaul codec network considers both historical fault monitoring data and current instantaneous state data simultaneously for equipment operation status analysis. This allows the network to capture the changing trend of equipment status over time, helping to discover potential faults or anomalies that are difficult to detect using only real-time data. At the same time, by combining historical fault monitoring data, the health status of the equipment can be assessed more comprehensively. In addition, the continuous codec structure ensures that misjudgments will not occur due to information loss or omission during feature analysis, thereby improving the accuracy of fault detection.

[0054] Example 2: Another embodiment of this application relates to an automatic equipment fault detection and early warning device. The implementation details of this embodiment's automatic equipment fault detection and early warning device are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution. A schematic diagram of this embodiment's automatic equipment fault detection and early warning device can be seen as follows: Figure 5 As shown, it includes a data acquisition module 301, a fault detection module 302, and an anomaly warning module 303.

[0055] The data acquisition module 301 is used to acquire current and historical fault monitoring data for the target port equipment. Specifically, the target port equipment includes dockside ship loaders, yard reclaimers, stacker-reclaimers, stackers, and other port equipment. The fault monitoring data includes damage monitoring time-series information for various types of port equipment. Time-series information refers to sequential data arranged according to the order of occurrence in time. Current fault monitoring data refers to the fault monitoring data at the current moment, and historical fault monitoring data includes at least the fault monitoring data from the previous moment.

[0056] In some examples, historical fault monitoring data includes fault monitoring data from n time points prior to the current time, where n ≥ 1, and the previous n time points are consecutive. Taking the scenario of structural damage monitoring for various types of port equipment as an example, fault monitoring data refers to the monitoring data obtained by monitoring structural damage to the target port equipment. The technologies used for structural damage monitoring include one of acoustic emission technology, ultrasonic guided wave technology, and strain response technology. The corresponding monitoring data are acoustic emission monitoring data, ultrasonic guided wave monitoring data, and strain monitoring data. These data are all time-series information, that is, sequential data arranged according to the order of occurrence in time. In practical applications, other technologies can also be used for structural damage monitoring of port equipment, not limited to the examples above.

[0057] In practical applications, this method is used to perform acoustic emission damage monitoring tasks (various defects may generate acoustic emission, including crack formation, plastic deformation, and phase transformation, etc. By detecting and analyzing the high-frequency sound waves emitted by the material when it is deformed or stressed through acoustic emission testing, corresponding fault monitoring data can be obtained). Before the structural damage fault occurs, that is, when the equipment is in normal condition, the acoustic emission signal detected by the acoustic emission sensor is a low-frequency noise signal, that is, it only contains background noise information; after the structural damage fault occurs, that is, when the equipment is in a fault state, the acoustic emission sensor detects structural damage, and at this time the acoustic emission signal contains the structural damage signal.

[0058] The fault detection module 302 is used to input the current fault monitoring data and historical fault monitoring data into the trained backhaul encoder-decoder network to obtain the fault detection result. Specifically, the backhaul encoder-decoder network includes an input encoding branch, a historical encoding branch, and a decoding branch. The input encoding branch is used to perform multi-level feature encoding on the current fault monitoring data, the historical encoding branch is used to generate historical feature embedding vectors corresponding to the historical fault monitoring data, and the decoding branch is used to concatenate the output of the input encoding branch and the historical feature embedding vectors and then perform multi-level feature decoding to obtain the fault detection result.

[0059] The anomaly warning module 303 is used to issue an early warning when the fault detection results indicate that the target port equipment is in an abnormal operating state.

[0060] This embodiment utilizes a backhaul codec network to simultaneously analyze the port equipment status based on current fault monitoring data and recent historical fault monitoring data, achieving comprehensive and reliable automatic fault detection and early warning for port equipment. Backhaul codec is achieved through a continuous codec structure and the reuse of historical fault monitoring data. By simultaneously considering both historical and current fault monitoring data in the port equipment operation status analysis, the codec network can capture the changing trends of the port equipment status over time, helping to discover potential faults or anomalies that are difficult to detect using only real-time fault monitoring data. Furthermore, combining historical fault monitoring data allows for a more comprehensive assessment of the equipment's health status. The continuous codec structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection.

[0061] In some embodiments, the input encoding branch includes at least three encoder levels to perform multi-level feature encoding on the current fault monitoring data, and the decoding branch includes at least three decoders to perform multi-level feature decoding by concatenating the output of the input encoding branch and historical feature embedding vector features. Furthermore, the feature encoding results output by each encoder level are fused during the multi-level feature decoding process. This continuous encoder-decoder structure ensures that misjudgments are not caused by information loss or omission during feature analysis, thereby improving the accuracy of fault detection. Fusing the feature encoding results output by each encoder level during the multi-level feature decoding process further enhances the accuracy of fault detection.

[0062] In some embodiments, the input encoding branch includes a three-level encoder, and the decoding branch includes a three-level decoder. The input encoding branch includes a first embedding module, a first encoder, a second encoder, and a third encoder; the historical encoding branch includes a second embedding module; and the decoding branch includes a first concatenation module, a first decoder, a second concatenation module, a second decoder, a third concatenation module, a third decoder, a fourth concatenation module, a linear layer, and a classification layer. The first embedding module is used to convert the current fault monitoring data into the current feature embedding vector and then input it into the first encoder. After multi-level feature encoding by the first encoder, the second encoder and the third encoder, the output is generated. The second embedding module is used to convert historical fault monitoring data into historical feature embedding vectors and then output them. The first concatenation module is used to concatenate the output of the input encoding branch and the historical feature embedding vector; The first decoder is used to decode the splicing result of the first splicing module. The second splicing module is used to splice the decoding result of the first decoder with the feature encoding result of the third encoder. The second decoder is used to decode the splicing result of the second splicing module. The third splicing module is used to splice the decoding result of the second decoder with the feature encoding result of the second encoder. The third decoder is used to decode the splicing result of the third splicing module. The fourth splicing module is used to splice the decoding result of the third decoder with the feature encoding result of the first encoder. The linear layer is used to perform a linear transformation on the splicing result of the fourth splicing module. The classification layer is used to output the fault detection result based on the linear transformation result.

[0063] In the actual implementation, all encoders and decoders use the same structure, and all feature splicing modules have the same internal structure.

[0064] In some embodiments, the first embedding module and the second embedding module generate word embedding vectors using the Word2vec word vector model, calculate a position encoding vector for each word embedding vector, and add the word embedding vector and the position encoding vector element by element to obtain the corresponding feature embedding vector. The Word2vec word vector model converts words in natural language into low-dimensional dense vectors that can be processed by computers, and allows the vectors to capture the semantic and grammatical relationships between words. For fault monitoring data of target port equipment, it is essentially the information expressed by words in natural language, such as the acoustic emission signal at time i. Using the Word2vec word vector model to convert the input acoustic emission signal at time i into a word embedding vector is a specific implementation of converting words in natural language into low-dimensional dense vectors that can be processed by computers. A word embedding vector is a type of low-dimensional dense vector that can be processed by computers.

[0065] The function of the first or second embedded module is to process the fault monitoring data output by the port equipment into the input format of the backhaul codec network in real time, while ensuring that the model can identify the fault information in the fault monitoring data of the port equipment.

[0066] In some examples, taking the first embedding module as an example, the current fault monitoring data is first converted into corresponding word embedding vectors. This step directly uses the Word2vec word vector model to generate word embedding vectors for the fault monitoring data. Then, a position encoding vector related to the position of each word embedding vector is calculated. The position encoding vector of each word embedding vector is calculated according to the following formula. Finally, the generated position encoding vector is added element by element to the input word embedding vector, thereby incorporating position information into the input and obtaining the final network input vector.

[0067] In the positional encoding vector calculation formula, the word position is p, the word embedding dimension is z, the model embedding dimension is d, and the even-dimensional positional embedding PE(p, 2z) and the odd-dimensional positional embedding PE(p, 2z+1) are calculated as follows:

[0068]

[0069] Here, the word position p refers to the position of the element in the current fault monitoring data sequence, and the word embedding dimension z and the embedding dimension d are both fixed values.

[0070] In some embodiments, the encoder and decoder include: Multi-head attention layer, used to extract multi-dimensional attention features from the input; The first normalization layer is used to normalize the sum of the output and input of the multi-head attention layer. The feedforward layer is used to perform a linear transformation on the normalized result. It uses a nonlinear function to activate the linear transformation result before performing the linear transformation. The second normalization layer is used to normalize the sum of the output and input of the feedforward layer.

[0071] In some examples, the encoder and decoder use the same structure, such as Figure 2 As shown. The input S of the codec is the output feature of the previous module, with a dimension equal to the model embedding dimension d. The codec construction process is as follows: First, the input S is fed into a multi-head attention layer for multi-dimensional attention feature extraction to fully extract feature information; then, the output of the multi-head attention layer is added to the input S itself, and the result is normalized by a normalization layer to obtain a normalized value D; next, the normalized value D is fed into a feedforward layer, which first performs a linear transformation on it, then activates it using the ReLU nonlinear function, and then performs another linear transformation. This step is to map the input normalized value D to a higher-dimensional space, then filters it using the ReLU nonlinear function, and finally transforms it back to the original dimension; the output of the previous step is added to the normalized value D itself, and the result is normalized by a normalization layer to obtain the final output F of the codec. Since the feature dimension is adjusted in the feedforward layer, the output F has the same dimension as the input S, which is d.

[0072] In some examples, the overall structure of a backhaul codec network is as follows: Figure 3 As shown, it mainly includes the following three steps: input encoding branch design, historical encoding branch design, and decoding branch design.

[0073] The input encoding branch structure is as follows: The fault monitoring data at the current moment is input into the first embedding module for preprocessing to generate the network input vector; the network input vector of dimension d is sequentially input into the first, second, and third encoders that are connected end to end to obtain the encoded outputs B1, B2, and B3 of dimension d, respectively. The historical encoding branch structure is as follows: historical fault monitoring data is input into the second embedding module for preprocessing, thereby generating historical feature embedding vectors; The decoding branch structure is as follows: The encoded output B3 of the third encoder and the historical feature embedding vector are input into the first feature concatenation module for feature concatenation. According to the embedding module, the embedding dimension is d. Therefore, in the first feature concatenation module, the encoded output B3 of the third encoder and the historical feature embedding vector are added together by dimension, and then normalized to output feature Q1 with dimension d.

[0074] Feature Q1 is input into the first decoder to obtain decoded feature J1; decoded feature J1 and encoded feature B3 from the third encoder are then input into the second feature concatenation module to obtain feature Q2 of dimension d; feature Q2 is input into the second decoder to obtain decoded feature J2; decoded feature J2 and encoded feature B2 from the second encoder are then input into the third feature concatenation module to obtain feature Q3 of dimension d; feature Q3 is input into the third decoder to obtain decoded feature J3; decoded feature J3 and encoded feature B1 from the first encoder are then input into the fourth feature concatenation module to obtain feature Q4 of dimension d.

[0075] Finally, the d-dimensional feature Q4 is output to the linear layer and then subjected to a fully connected layer linear transformation, resulting in an output dimension of 2, which is then output to the classification layer. In this embodiment, the classification layer uses the softmax function, and the output is 2-dimensional data, representing the probabilities of the input data being normal and abnormal data, respectively.

[0076] The above describes the construction process of a return-to-work encoder-decoder network. The input to the encoder and decoder is the output of the previous module, and the output of each encoder and decoder is its encoded result.

[0077] In some embodiments, the above-mentioned automatic equipment fault detection and early warning device further includes: a network training module, used for: conducting a simulated acquisition experiment of fault monitoring data of the target port equipment, debugging different states of the target port equipment, recording fault monitoring data before and after the fault occurs in the fault state, and fault monitoring data in the non-fault state, determining the fault monitoring data before and after the fault occurs as a fault data group, determining the fault monitoring data in the non-fault state as a normal data group, the fault data group and the normal data group constitute a fault detection dataset; and training a backhaul encoding and decoding network based on the fault detection dataset.

[0078] Specifically, fault monitoring data simulation experiments are conducted to determine different fault states. Fault monitoring data before the fault occurs is recorded as historical fault monitoring data, and fault monitoring data after the fault occurs is recorded as current fault monitoring data. The pairs of historical and current fault monitoring data are labeled as fault data groups. In addition, a large number of historical and current fault monitoring data pairs are recorded under normal conditions of the port equipment and labeled as normal data groups. After obtaining a sufficient number of fault and normal data groups, a fault detection dataset is constructed. In some examples, "sufficient number" can refer to a sample size of over 10,000, enough to support backhaul encoding / decoding network training until the loss function converges.

[0079] In practice, erosion from the climate and long-term operational wear can cause corrosion, cracks, and plastic deformation in port equipment, leading to performance degradation. Fault signal acquisition methods are divided into two types: 1. Normal data acquisition on real, intact equipment, and fault data acquisition on existing, damaged equipment; 2. Fabricating a large number of simulated structural components, artificially creating corrosion, cracks, and plastic deformation on these components to achieve fault monitoring data acquisition before and after a fault occurs. In this embodiment, a fault monitoring data simulation acquisition experiment is conducted, artificially creating corrosion, cracks, and plastic deformation on simulated structural components to achieve various fault states such as corrosion, cracks, and plastic deformation, thereby enabling fault monitoring data acquisition before and after a fault occurs.

[0080] In some examples, the backhaul codec network is trained on the fault detection dataset. During the training process, the ratio of fault data group to normal data group is set to 4:1 to ensure that the backhaul codec network pays more attention to the characteristics of fault data. After the training converges, the final backhaul codec network is obtained.

[0081] In practice, the construction process of a backhaul codec network is as follows: Figure 4 As shown, firstly, a fault detection dataset is constructed; then, a device signal embedding module, a codec, and a backhaul codec network are designed, and finally, a backhaul codec network is generated.

[0082] In some embodiments, the fault detection result includes a first score and a second score. When the first score exceeds a preset value, it indicates that the target port equipment is in an abnormal operating state.

[0083] In the specific implementation, the first score and the second score are the abnormal data score and the normal data score, respectively. In practical applications, the abnormal data score and the normal data score can represent the probability of a fault occurring and the probability of no fault occurring, respectively. A backhaul codec model is used to monitor the equipment status in real time. The fault monitoring data of the port equipment is input into the backhaul codec network in real time, and the corresponding abnormal data score and normal data score are output. This allows us to obtain the probability of the target port equipment malfunctioning and the probability of no fault occurring. When the abnormal data score is greater than a preset value (for example, it is set to 0.5 based on experience, that is, the probability of a fault exceeding 50%, then an alarm is triggered; in practical applications, this preset value can also be manually adjusted according to the characteristics of the dataset and experimental results), an alarm is triggered, indicating that the current port equipment may be in an abnormal operating state.

[0084] This embodiment of the device for automatic equipment fault detection and early warning based on a backhaul codec network analyzes equipment status simultaneously based on current fault monitoring data and recent historical fault monitoring data. This enables automatic monitoring and fault early warning of intelligent equipment, saving labor costs, improving operational efficiency, reducing risks, and supporting the sustainable development of smart port construction. The beneficial effects of this embodiment are: (1) Compared with the 24 / 7 human-managed operation and maintenance method, the backhaul codec network can monitor the equipment status in real time without human intervention and provide real-time early warning of potential risks in equipment operation. It can significantly save manpower costs, improve operation and maintenance efficiency, and effectively ensure the safe and efficient operation of port equipment.

[0085] (2) Compared to ordinary fault detection networks, which only use a single current instantaneous state output for analysis, the backhaul codec network considers both historical fault monitoring data and current instantaneous state data simultaneously for equipment operation status analysis. This allows the network to capture the changing trend of equipment status over time, helping to discover potential faults or anomalies that are difficult to detect using only real-time data. At the same time, by combining historical fault monitoring data, the health status of the equipment can be assessed more comprehensively. In addition, the continuous codec structure ensures that misjudgments will not occur due to information loss or omission during feature analysis, thereby improving the accuracy of fault detection.

[0086] Example 3: Another embodiment of this application relates to an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the automatic detection and early warning methods for device faults described in the above embodiments.

[0087] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0088] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0089] Example 4: Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the above-described embodiment of the automatic device fault detection and early warning method.

[0090] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for automatic detection and early warning of equipment faults, characterized in that, include: Acquire current and historical fault monitoring data for the target port equipment; The current fault monitoring data and historical fault monitoring data are input into the trained backhaul encoder-decoder network to obtain the fault detection results. An early warning is issued when fault detection results indicate that the target port equipment is in an abnormal operating state; The backhaul encoding / decoding network includes an input encoding branch, a history encoding branch, and a decoding branch. The input encoding branch is used to perform multi-level feature encoding on the current fault monitoring data. The history encoding branch is used to generate historical feature embedding vectors corresponding to historical fault monitoring data. The decoding branch is used to concatenate the output of the input encoding branch and the historical feature embedding vectors and then perform multi-level feature decoding to obtain the fault detection result.

2. The automatic equipment fault detection and early warning method according to claim 1, characterized in that, The input encoding branch includes at least three encoders to perform multi-level feature encoding on the current fault monitoring data. The decoding branch includes at least three decoders to perform multi-level feature decoding by concatenating the output of the input encoding branch and the historical feature embedding vector features. Furthermore, the feature encoding results output by each level encoder are fused during the multi-level feature decoding process.

3. The automatic equipment fault detection and early warning method according to claim 1, characterized in that, The input encoding branch includes a three-level encoder, and the decoding branch includes a three-level decoder; the input encoding branch includes a first embedding module, a first encoder, a second encoder, and a third encoder, the historical encoding branch includes a second embedding module, and the decoding branch includes a first splicing module, a first decoder, a second splicing module, a second decoder, a third splicing module, a third decoder, a fourth splicing module, a linear layer, and a classification layer; The first embedding module is used to convert the current fault monitoring data into the current feature embedding vector and then input it into the first encoder. After multi-level feature encoding by the first encoder, the second encoder and the third encoder, the output is generated. The second embedding module is used to convert historical fault monitoring data into historical feature embedding vectors and then output them. The first concatenation module is used to concatenate the output of the input encoding branch and the historical feature embedding vector; The first decoder is used to decode the splicing result of the first splicing module. The second splicing module is used to splice the decoding result of the first decoder with the feature encoding result of the third encoder. The second decoder is used to decode the splicing result of the second splicing module. The third splicing module is used to splice the decoding result of the second decoder with the feature encoding result of the second encoder. The third decoder is used to decode the splicing result of the third splicing module. The fourth splicing module is used to splice the decoding result of the third decoder with the feature encoding result of the first encoder. The linear layer is used to perform a linear transformation on the splicing result of the fourth splicing module. The classification layer is used to output the fault detection result based on the linear transformation result.

4. The automatic equipment fault detection and early warning method according to claim 1, characterized in that, Also includes: Conduct a simulation and acquisition experiment of fault monitoring data of target port equipment, debug different states of target port equipment, record fault monitoring data before and after the fault occurs in the fault state, and fault monitoring data in the non-fault state. The fault monitoring data before and after the fault occurs is determined as the fault data group, and the fault monitoring data in the non-fault state is determined as the normal data group. The fault data group and the normal data group constitute the fault detection dataset. The backhaul encoder-decoder network is trained based on a fault detection dataset.

5. The automatic equipment fault detection and early warning method according to claim 1, characterized in that, The fault detection results include a first score and a second score. When the first score exceeds a preset value, it indicates that the target port equipment is in an abnormal operating state.

6. The automatic equipment fault detection and early warning method according to claim 2, characterized in that, The encoder and the decoder adopt the same structure, including: Multi-head attention layer, used to extract multi-dimensional attention features from the input; The first normalization layer is used to normalize the sum of the output and input of the multi-head attention layer. The feedforward layer is used to perform a linear transformation on the normalized result. It uses a nonlinear function to activate the linear transformation result before performing the linear transformation. The second normalization layer is used to normalize the sum of the output and input of the feedforward layer.

7. The automatic equipment fault detection and early warning method according to claim 3, characterized in that, The first embedding module and the second embedding module generate word embedding vectors using the Word2vec word vector model, calculate position encoding vectors for each word embedding vector, and add the word embedding vectors and position encoding vectors element by element to obtain the corresponding feature embedding vectors.

8. An automatic equipment fault detection and early warning device, characterized in that, include: The data acquisition module is used to acquire current and historical fault monitoring data for the target port equipment. The fault detection module is used to input current fault monitoring data and historical fault monitoring data into the trained backhaul encoder-decoder network to obtain fault detection results; The anomaly warning module is used to issue an early warning when the fault detection results indicate that the target port equipment is in an abnormal operating state; The backhaul encoding / decoding network includes an input encoding branch, a history encoding branch, and a decoding branch. The input encoding branch is used to perform multi-level feature encoding on the current fault monitoring data. The history encoding branch is used to generate historical feature embedding vectors corresponding to historical fault monitoring data. The decoding branch is used to concatenate the output of the input encoding branch and the historical feature embedding vectors and then perform multi-level feature decoding to obtain the fault detection result.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the automatic detection and early warning method for equipment faults as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the automatic detection and early warning method for equipment faults as described in any one of claims 1 to 7.

Images