Artificial intelligence-based device fault identification method and system, and storage medium

Abstract

The application discloses a kind of based on artificial intelligence equipment fault identification method, system and storage medium, belong to equipment fault identification technical field, including: monitoring equipment based on mode category acquisition any agricultural equipment state monitoring data, based on mode category respectively for any state monitoring data in historical time series respectively mark fault label or non-fault label;Construct fault prediction model, fault prediction model output agricultural equipment in prediction time analysis sequence, and based on analysis sequence output analysis prompt information;Based on equipment type, construct fault knowledge base, based on analysis prompt information judge whether there is fault equipment in all agricultural equipment, if there is, then output the fault category of fault equipment, and extract the fault analysis list corresponding to fault category in fault knowledge base;Predictive time, fault category and fault analysis list are sent to management personnel. Through the application, the accuracy and efficiency of equipment fault identification can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of equipment fault identification, and specifically relates to an equipment fault identification method, system and storage medium based on artificial intelligence. Background Art

[0002] With the rapid development of industrial automation and informatization, the complexity of various production equipment and systems continues to increase. This is particularly true in the agricultural and animal husbandry industries, where agricultural and animal husbandry equipment is prone to various failures during long-term operation. With the diversification of equipment types and the complexity of failure modes, traditional fault diagnosis methods require extensive expertise and experience, which limits their widespread application and adaptability. Therefore, by leveraging artificial intelligence technology, fault signatures can be automatically learned from equipment operating data, enabling real-time monitoring of equipment status and early warning of faults.

[0003] Similar prior art includes Chinese patent application publication number CN116150676A, which discloses an artificial intelligence-based equipment fault diagnosis and identification method and apparatus. This relates to the field of fault diagnosis technology and includes a training method and apparatus for an equipment fault diagnosis model, as well as an equipment fault diagnosis method and apparatus. The training method for the equipment fault diagnosis model includes: obtaining a pre-collected signal data set; the signal data set includes a first characteristic parameter representing the normal operating state of the equipment and a second characteristic parameter representing multiple abnormal states; performing feature extraction on the first characteristic parameter and the second characteristic parameter to obtain target characteristic parameters respectively included in the first characteristic parameter and the second characteristic parameter; inputting the target characteristic parameters into a pre-set initial classifier, training the initial classifier, and obtaining an equipment fault diagnosis model. The equipment fault diagnosis model corresponding to this invention can effectively improve the accuracy of equipment fault identification in equipment fault diagnosis tasks. There is also Chinese patent application publication number CN113689111A, which discloses a training method and apparatus for a fault identification model, a fault identification method and apparatus, an electronic device, and a medium. These relate to artificial intelligence, particularly the field of deep learning. A training method for a fault identification model includes obtaining a first fault data set; generating a first sample set based on the first fault data set, wherein each sample in the first sample set includes a fault description text and a corresponding fault component and a fault type; and using the fault description text of the sample in the first sample set as input to the fault identification model, and adjusting the parameters of the fault identification model so that the fault identification model outputs a first label and a second label, wherein the first label is a fault component corresponding to the input fault description text, and the second label is a fault type corresponding to the fault component.

[0004] However, the above-mentioned existing technologies only use target characteristic parameters to diagnose equipment faults, and the fault identification model cannot accurately identify fault types outside the first sample set. In actual situations, equipment faults involve linkage factors between components, and it is necessary to combine multiple status monitoring data for equipment fault identification. Summary of the Invention

[0005] In response to the technical problems raised above, the present invention provides an artificial intelligence-based equipment fault identification method, system and storage medium to solve the problems in the prior art.

[0006] The present invention proposes an artificial intelligence-based equipment fault identification method, comprising:

[0007] Obtaining the equipment type of any agricultural and animal husbandry equipment, using a monitoring device to collect status monitoring data of any agricultural and animal husbandry equipment based on a pattern category, organizing the status monitoring data corresponding to any pattern category into a historical time series based on a collection time, and marking any status monitoring data in the historical time series with a fault label or a non-fault label based on the pattern category;

[0008] Building a fault prediction model based on all historical time series, the fault prediction model outputs an analysis sequence of the agricultural and animal husbandry equipment within a prediction time, and outputs analysis prompt information based on the analysis sequence;

[0009] Building a fault knowledge base based on the equipment type, determining whether there is a faulty device among all the agricultural and animal husbandry equipment based on the analysis prompt information, and if so, outputting the fault category of the faulty device, and extracting a fault analysis list corresponding to the fault category from the fault knowledge base;

[0010] The predicted time, the fault category and the fault analysis list are sent to a management personnel.

[0011] Furthermore, building a fault prediction model based on all historical time series includes the following steps:

[0012] Constructing a hybrid prediction model based on a neural network, combining all historical time series based on the acquisition time to generate a historical combination sequence, dividing the historical combination sequence into multiple subsequence combinations based on a clustering algorithm, sequentially calculating the slopes of the subsequence combinations between adjacent acquisition times, combining all the slopes and setting them as feature quantities corresponding to the subsequence combinations;

[0013] All the feature quantities are input into the hybrid prediction model, and the feature scores corresponding to the historical combination sequence are calculated based on the feature quantities. If there is a feature score greater than a first threshold, the hybrid prediction model is trained based on the feature quantities, and the trained hybrid prediction model is set as the fault prediction model. Otherwise, this step is repeated after adjusting the parameters of the clustering algorithm until the feature score is greater than the first threshold.

[0014] Furthermore, calculating the feature score based on the feature quantity includes:

[0015] Obtaining respectively the distribution ratios of the fault labels and the non-fault labels corresponding to all the feature quantities in the subsequence combination, and the identification ratios of the fault labels and the non-fault labels corresponding to all the feature quantities identified by the hybrid prediction model, and setting the product of the distribution ratio and the identification ratio corresponding to the same label as a first value;

[0016] The largest first value is extracted from any subsequence combination and set as the second value, and all the second values ​​contained in the historical combination sequence are accumulated and set as the feature score.

[0017] Furthermore, the analysis sequence and the analysis prompt information are output based on the following steps:

[0018] The hybrid prediction model includes a deep neural network and a long short-term memory network, wherein the middle layer of the deep neural network models the feature quantity corresponding to any of the subsequence combinations through a mixed Gaussian distribution, and outputs a state estimate corresponding to any of the subsequence combinations based on the pattern category;

[0019] The long short-term memory network outputs a change sequence of the historical combination sequence within the prediction time based on all the feature quantities, obtains a distribution probability of the change sequence in all the state estimates based on the mixed Gaussian distribution, and extracts the state estimate corresponding to the change sequence based on the distribution probability and sets it as the state type;

[0020] The change sequence is set as the analysis sequence, a device ID is marked for the analysis sequence based on the device type, and the device ID corresponding to the analysis sequence and the state type are combined to generate the analysis prompt information.

[0021] Furthermore, multiple threshold intervals are set based on the mode category and the monitoring equipment. If there is sequence data in the analysis sequence that does not belong to the threshold interval, the agricultural and animal husbandry equipment corresponding to the equipment ID is set as a faulty equipment, and the fault category corresponding to the sequence data is obtained based on the status type.

[0022] Furthermore, building a fault knowledge base based on the device type includes the following steps:

[0023] Collecting equipment manuals and historical fault records of any of the agricultural and animal husbandry equipment based on the same equipment type, and using the equipment manuals and historical fault records as reference records;

[0024] Building a text recognition model based on natural language technology, setting component labels based on the device manual, extracting component failure information and processing information corresponding to the component labels from the reference records based on part-of-speech categories, and matching the component failure information with the processing information based on distribution distance;

[0025] Setting any of the agricultural and animal husbandry equipment as a parent node, setting the component label as a child node, setting the component fault information as a phenomenon node, setting the matched processing information as an analysis node, performing semantic expansion on the component fault information to generate extended fault information, adding the extended fault information as an extension node in the phenomenon node, marking component connection relationships in the child nodes based on the equipment manual, and sequentially performing directed connections on the parent node, the child node, the phenomenon node, the extension node, the analysis node, and the component connection relationships to generate a knowledge distribution table;

[0026] The knowledge distribution tables corresponding to all the agricultural and animal husbandry equipment are combined to generate the fault knowledge base, and the phenomenon nodes, the extension nodes and the sub-nodes are directed connected based on a preset format to generate a connection formula corresponding to the knowledge distribution table.

[0027] Furthermore, matching the component fault information with the processing information based on the distribution distance includes the following steps:

[0028] In the reference record, the sequential occurrence frequencies of any component failure information and any processing information are obtained based on a time relationship, the sequential occurrence frequencies are combined into multiple distribution vectors, and the distribution distance D is calculated based on a first formula, which is: Wherein, x1 is the frequency of the sequential occurrence of the processing information before the component fault information in the distribution vector, x2 is the frequency of the sequential occurrence of the processing information after the component fault information in the distribution vector, y1 is the frequency of the sequential occurrence of the component fault information before the processing information in the distribution vector, and y2 is the frequency of the sequential occurrence of the component fault information after the processing information in the distribution vector;

[0029] If the distribution distance is less than or equal to a second threshold, the processing information and the component failure information are determined to be matched.

[0030] Furthermore, extracting the fault analysis list includes the following steps:

[0031] Extracting device fault information from the fault category based on the preset format, calculating semantic similarity between the device fault information and the phenomenon node and the extension node in any of the connection formulas, extracting the connection formulas with semantic similarity greater than a third threshold and setting them as a set of candidate connection formulas;

[0032] Extracting the connection type corresponding to the agricultural and animal husbandry equipment from the set of candidate connection types based on the analysis prompt information and setting it as a suspected fault;

[0033] The suspected fault is located in the fault knowledge base, node information corresponding to the suspected fault is obtained based on the knowledge distribution table, and the node information is summarized to generate the fault analysis list.

[0034] The present invention also provides an artificial intelligence-based equipment fault identification system, which is used to implement the above-mentioned artificial intelligence-based equipment fault identification method. The system mainly includes:

[0035] An acquisition module obtains the equipment type of any agricultural and animal husbandry equipment, and a monitoring device collects status monitoring data of any agricultural and animal husbandry equipment based on a pattern category, organizes the status monitoring data corresponding to any pattern category into a historical time series based on the collection time, and labels any status monitoring data in the historical time series with a fault label or a non-fault label based on the pattern category;

[0036] A prediction module constructs a fault prediction model based on all historical time series, the fault prediction model outputs an analysis sequence of the agricultural and animal husbandry equipment within the prediction time, and outputs analysis prompt information based on the analysis sequence;

[0037] an identification module, which builds a fault knowledge base based on the equipment type, determines whether there is a faulty device among all the agricultural and animal husbandry equipment based on the analysis prompt information, and if so, outputs the fault category of the faulty device and extracts a fault analysis list corresponding to the fault category from the fault knowledge base;

[0038] A sending module sends the predicted time, the fault category and the fault analysis list to a management personnel.

[0039] The present invention also provides a computer storage medium storing program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the above-mentioned artificial intelligence-based device fault identification method.

[0040] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0041] The present invention first forms a historical time series by marking the labels of the status monitoring data, and trains and generates a fault prediction model, which can accurately verify the prediction accuracy of the fault prediction model. Then, a fault knowledge base is constructed through the knowledge distribution table, which can improve the comprehensiveness of equipment fault analysis. Finally, the text recognition model filters out the node information in the knowledge distribution table through the semantic similarity between each node in the connection formula and the equipment fault information, and generates a fault analysis list. It can not only accurately identify the fault category of agricultural and animal husbandry equipment, but also quickly generate solutions through the fault analysis list, thereby achieving high efficiency in solving equipment faults.

[0042] The present invention sets the feature quantity by the slope corresponding to the subsequence combination, which can reduce the complexity of the hybrid prediction model in performing feature analysis on all subsequence combinations. It also accurately matches the component fault information and processing information in the reference record through the distribution distance, which can improve the integrity of the fault knowledge base. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flowchart of the steps of an artificial intelligence-based equipment fault identification method of the present invention;

[0044] Figure 2 A schematic diagram of generating a knowledge distribution table in the present invention;

[0045] Figure 3 This is a structural diagram of an equipment fault identification system based on artificial intelligence in the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] It is understood that the terms "first," "second," etc., used herein may be used to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script without departing from the scope of this application.

[0048] like Figure 1 As shown, an artificial intelligence-based equipment fault identification method includes:

[0049] S1: Obtain the equipment type of any agricultural and animal husbandry equipment. The monitoring equipment collects the status monitoring data of any agricultural and animal husbandry equipment based on the pattern category. The status monitoring data corresponding to any pattern category are organized into a historical time series based on the collection time. Based on the pattern category, any status monitoring data in the historical time series is marked with a fault label or a non-fault label.

[0050] Specifically, in this embodiment, agricultural and animal husbandry equipment refers to various mechanical equipment and tools used in agricultural production (including planting, animal husbandry, and fishery) and animal husbandry production (including grazing and breeding), such as planting equipment, animal husbandry equipment, fishery equipment, and facility agriculture equipment. Equipment type refers to the type information corresponding to agricultural and animal husbandry equipment, including but not limited to mechanical equipment, breeding equipment, farming equipment, and greenhouse equipment. Mode category refers to the operating mode of the equipment during operation, such as fault operation mode, low-speed operation mode, and high-frequency operation mode. Different mode categories represent different operating modes of agricultural and animal husbandry equipment. Monitoring equipment refers to various types of sensors. Status monitoring data refers to data collected by monitoring the operation of agricultural and animal husbandry equipment using various types of sensors. Therefore, status monitoring data includes multiple types, such as temperature, vibration frequency, pressure data, and load consumption data. Acquisition time refers to the timestamp corresponding to the time when the sensor generates status monitoring data. Historical time series refers to a sequence of status monitoring data corresponding to any sensor type, formed by combining acquisition times, which contains multiple discrete data. The fault label refers to the label information corresponding to the data collected when the agricultural and animal husbandry equipment is in a fault operation mode, and the non-fault label refers to the label information corresponding to the data collected when the agricultural and animal husbandry equipment is in a non-fault operation mode.

[0051] S2: Build a fault prediction model based on all historical time series. The fault prediction model outputs the analysis sequence of agricultural and animal husbandry equipment within the prediction time, and outputs analysis prompt information based on the analysis sequence.

[0052] Specifically, in this embodiment, by labeling historical time series, the prediction accuracy of the fault prediction model can be improved. The prediction time refers to the future monitoring period, and the analysis sequence refers to the sequence of data corresponding to each timestamp within the prediction time. The analysis prompt information refers to the information generated by the analysis sequence to estimate and analyze the operating status of agricultural and animal husbandry equipment.

[0053] S3: Build a fault knowledge base based on the equipment type, and determine whether there is a faulty device among all agricultural and animal husbandry equipment based on the analysis prompt information. If so, output the fault category of the faulty device and extract the fault analysis list corresponding to the fault category from the fault knowledge base.

[0054] Specifically, in this embodiment, the fault knowledge base refers to a database of device knowledge covering all device types, enabling rapid retrieval of device faults, analysis of causes, and troubleshooting procedures. The fault category describes the current operating mode of the faulty device, for example, intermittent operation due to low temperature. The fault analysis list includes information such as the faulty component, fault type, cause, and troubleshooting instructions for the faulty device.

[0055] S4: Send the predicted time, fault category and fault analysis list to the management personnel.

[0056] Specifically, through the above steps, agricultural and animal husbandry equipment can be monitored in real time, and equipment failures can be accurately identified within the predicted time. At the same time, the predicted time, failure category and failure analysis list are sent to the management personnel, which can improve the maintenance efficiency of the faulty parts.

[0057] As a preferred technical solution of the present invention, building a fault prediction model based on all historical time series includes the following steps:

[0058] A hybrid prediction model is constructed based on a neural network. All historical time series are combined based on the collection time to generate a historical combination sequence. The historical combination sequence is divided into multiple subsequence combinations based on the clustering algorithm. The slopes of the subsequence combinations between adjacent collection times are calculated in sequence. All slopes are combined and set as the feature quantities corresponding to the subsequence combinations.

[0059] All feature quantities are input into the hybrid prediction model, and the feature scores corresponding to the historical combination sequences are calculated based on the feature quantities. If there is a feature score greater than the first threshold, the hybrid prediction model is trained based on the feature quantities, and the trained hybrid prediction model is set as the fault prediction model. Otherwise, the parameters of the clustering algorithm are adjusted and this step is repeated until the feature score is greater than the first threshold.

[0060] Specifically, in this embodiment, the hybrid prediction model is a model generated by combining multiple neural networks, including a deep neural network (DNN) and a long short-term memory network (LSTM), wherein a deep neural network refers to a neural network with multiple hidden layers, and a long short-term memory network is also a type of deep neural network. Since there are multiple historical time series, all historical time series are combined to generate a historical combination sequence through the same acquisition time, which can monitor the correlation between the operating data of agricultural and animal husbandry equipment in multiple directions. The clustering algorithm is an unsupervised learning method that can group sample data in the historical combination sequence according to similarity, including but not limited to clustering algorithms such as K-means and hierarchical clustering. The slope is used to calculate the rate of change between the discrete data corresponding to any subsequence in the subsequence combination, and the slope of any subsequence is combined in sequence. Then, a multidimensional data matrix is ​​generated by the number of subsequences in the subsequence combination, that is, the feature quantity. For example, the subsequence combination D contains two subsequences D1 and D2, where the slope corresponding to subsequence D1 is (0.4, 0.7, 0.7) and the slope corresponding to subsequence D2 is (0.1, 0.8, 0.1). Then the feature quantity corresponding to the subsequence combination D is

[0061] The feature score is used to measure the clustering effectiveness of each sub-sequence combination. If the feature score is greater than the first threshold, it indicates that the clustering effectiveness of the corresponding sub-sequence combination is good and the hybrid prediction model has a high accuracy rate in identifying feature quantities. Therefore, the feature quantities can be used to train and learn the hybrid prediction model. Otherwise, the clustering algorithm parameters need to be readjusted to re-divide the multiple sub-sequence combinations. This step is repeated until the feature score of the feature quantity corresponding to the sub-sequence combination is greater than the first threshold. This step can improve the training accuracy of the hybrid prediction model for all feature quantities. Therefore, the trained hybrid prediction model is set as the fault prediction model.

[0062] As a preferred technical solution of the present invention, calculating the feature score based on the feature quantity includes:

[0063] The distribution ratios of fault labels and non-fault labels corresponding to all feature quantities in the subsequence combination and the identification ratios of fault labels and non-fault labels corresponding to all feature quantities identified by the hybrid prediction model are obtained respectively, and the product of the distribution ratio and the identification ratio corresponding to the same label is set as a first value.

[0064] The largest first value is extracted from any subsequence combination and set as the second value. All second values ​​contained in the historical combination sequence are accumulated and set as the feature score.

[0065] Specifically, in this embodiment, since the features corresponding to all subsequence combinations are trained using labels, the recognition accuracy of the hybrid prediction model needs to be verified using labels. All features in any subsequence combination are represented by a multidimensional matrix. The distribution ratio refers to the proportion of features that actually belong to faulty labels versus non-faulty labels. For example, 30% of the features belong to faulty labels, while 70% belong to non-faulty labels. The recognition ratio refers to the accuracy with which the hybrid prediction model can distinguish features with faulty and non-faulty labels. For example, in subsequence combination M1, the accuracy for identifying features with faulty labels is 80%, while the accuracy for identifying features with non-faulty labels is 90%. Identical labels refer to either the same faulty label or the same non-faulty label. For example, the first value corresponding to the faulty label is 30% × 80% = 0.24, while the first value corresponding to the non-faulty label is 70% × 90% = 0.63.

[0066] For example, if the largest first value is extracted, the second value corresponding to the subsequence combination M1 is 0.63. Since the historical combination sequence is split and clustered into multiple subsequence combinations, the feature score can be calculated by accumulating the second values ​​corresponding to all subsequence combinations.

[0067] As a preferred technical solution of the present invention, the analysis sequence and analysis prompt information are output based on the following steps:

[0068] The hybrid prediction model includes a deep neural network and a long short-term memory network. The middle layer of the deep neural network models the feature quantities corresponding to any subsequence combination through a mixed Gaussian distribution, and outputs the state estimation corresponding to any subsequence combination based on the pattern category.

[0069] The long short-term memory network outputs the change sequence of the historical combination sequence within the prediction time based on all feature quantities, obtains the distribution probability of the change sequence in all state estimates based on the mixed Gaussian distribution, and extracts the state estimate corresponding to the change sequence based on the distribution probability and sets it as the state type.

[0070] The change sequence is set as the analysis sequence, the device ID is marked for the analysis sequence based on the device type, and the device ID and status type corresponding to the analysis sequence are combined to generate analysis prompt information.

[0071] Specifically, in this embodiment, the deep neural network generally includes an input layer, an intermediate layer, and an output layer. The intermediate layer is used to identify and distinguish all feature quantities, which helps to distinguish the features of various states. Once the features are extracted from the deep neural network, all the features can be used to train a mixed Gaussian distribution model through the output layer, wherein the mixed Gaussian distribution (Gaussian Mixture Model, GMM) is a probability-based machine learning algorithm. For example, the mixed Gaussian distribution learns the feature distribution of various feature quantities, and models based on the feature distribution. The feature distribution can be estimated in state according to the pattern category, wherein the feature distribution is composed of features generated after the intermediate layer of the deep neural network extracts any feature quantity. The generated features usually include an abstract representation of the input feature quantity. There are multiple state estimates, and the state estimates include state information corresponding to the pattern category and information about monitoring equipment with abnormal data in the subsequence combination.

[0072] A change sequence is the combined result of using a long short-term memory network to predict any sequence in a historical combination sequence. The long short-term memory network can analyze and predict the segmentation steps of each subsequence combination in the historical combination sequence, which can improve the accuracy of the change sequence. For example, the long short-term memory network obtains time series features corresponding to features extracted by a deep neural network, that is, the output of the DNN can be directly connected to the input of the LSTM. The distribution probability corresponding to each feature distribution in the mixed Gaussian distribution of the change sequence can then be used to extract the state type corresponding to the change sequence, such as a low-speed operation state caused by low temperature.

[0073] The device ID refers to the serial number of each agricultural and animal husbandry equipment. Combining the device ID with the status type can accurately describe whether there is a fault in the analysis sequence, which is the analysis prompt information.

[0074] As a preferred technical solution of the present invention, multiple threshold intervals are set based on the pattern category and the monitoring equipment. If there is sequence data in the analysis sequence that does not belong to the threshold interval, the agricultural and animal husbandry equipment corresponding to the equipment ID is set as a faulty equipment, and the fault category corresponding to the sequence data is obtained based on the status type.

[0075] Specifically, the threshold interval refers to the numerical range corresponding to each monitoring device when the agricultural and animal husbandry equipment is operating normally under different mode categories. The threshold interval can be used to determine whether the analysis sequence is operating normally. The status category can be used to filter out the category of abnormal data of the faulty equipment and set it as the fault category. Among them, equipment fault identification can be achieved through the fault category.

[0076] As a preferred technical solution of the present invention, building a fault knowledge base based on device type includes the following steps:

[0077] Based on the same equipment type, the equipment manual and historical fault records of any agricultural and animal husbandry equipment are collected, and the equipment manual and historical fault records are used as reference records.

[0078] A text recognition model is built based on natural language technology, and component labels are set based on the equipment manual. The text recognition model extracts component failure information and processing information corresponding to the component labels in reference records based on part-of-speech categories, and matches the component failure information with the processing information based on distribution distance.

[0079] Set any agricultural and animal husbandry equipment as the parent node, set the component label as the child node, set the component fault information as the phenomenon node, set the matched processing information as the analysis node, perform semantic expansion on the component fault information to generate extended fault information, add the extended fault information as an extension node in the phenomenon node, mark the component connection relationship in the child node based on the equipment manual, and connect the parent node, child node, phenomenon node, extension node, analysis node and component connection relationship in a directed manner in turn to generate a knowledge distribution table.

[0080] The knowledge distribution tables corresponding to all agricultural and animal husbandry equipment are combined to generate a fault knowledge base, and the phenomenon nodes, extension nodes and sub-nodes are connected in a directed manner based on the preset format to generate the connection formula corresponding to the knowledge distribution table.

[0081] Specifically, in this embodiment, in order to build an accurate fault knowledge base, it is necessary to collect equipment manuals and historical fault records corresponding to agricultural and animal husbandry equipment of the same equipment type, where the equipment manual includes but is not limited to information such as the structural diagram, name and working principle of each component, and historical fault records refer to information recording the maintenance and repair of agricultural and animal husbandry equipment, which are summarized to generate reference records.

[0082] Natural Language Processing (NLP) technology is a branch of artificial intelligence and linguistics, applicable to text analysis, classification, and relationship extraction. Therefore, the present invention uses NLP technology to construct a text recognition model. A component label refers to the label information corresponding to a component name. Typically, agricultural and animal husbandry equipment is composed of multiple component names. A part-of-speech category refers to the type of part of speech, including but not limited to nouns, verbs, and adjectives. Component failure information describes the phenomenon when a component label fails, and processing information refers to the actions taken to address the failure corresponding to the component label. For example, "The vibration frequency of component F1 is 100 Hz, exceeding the threshold range; adjust the bearing," where "component F1" is the component label, "The vibration frequency is 100 Hz, exceeding the threshold range" is the component failure information, and "Adjust the bearing" is the processing information. Distribution distance refers to a numerical value used in natural language processing to measure the difference between two probability distributions. It can be used to match component failure information and processing information in reference records to ensure a one-to-one correspondence. The smaller the distribution distance, the smaller the difference, and the closer the match between the corresponding component failure information and processing information.

[0083] Since component fault information may be incomplete, thus affecting the accuracy of subsequent fault identification, it is necessary to semantically expand component fault information through natural language processing technology to generate extended fault information. Figure 2 As shown, by setting multiple nodes and generating a knowledge distribution table H1 after directed connection, the comprehensiveness and relevance of the fault knowledge base can be improved. Among them, the component connection relationship refers to the operating connection relationship between the various components in the agricultural and animal husbandry equipment, such as series relationship, sequence relationship and parallel relationship, etc. Directed connection refers to a directional connection method.

[0084] In order to quickly identify faults later, the knowledge distribution table can be described in a connected manner to reduce the text processing process, such as Figure 2 As shown, the connection formula corresponding to the knowledge distribution table H1 is h1, where the preset format refers to entity nouns containing time sequence. For example, the information corresponding to the phenomenon node is "temperature 100, speed 10", then "temperature 100" appears before "speed 10", which may be because the temperature is too high, resulting in a decrease in speed. Therefore, it is split into two phenomenon nodes, and the directed connection is: "temperature 100" → "speed 10", and both are connected to the corresponding child nodes.

[0085] As a preferred technical solution of the present invention, matching component fault information with processing information based on distribution distance includes the following steps:

[0086] In the reference record, the sequential occurrence frequency of any component fault information and any processing information is obtained based on the time relationship, the sequential occurrence frequencies are combined into multiple distribution vectors, and the distribution distance D is calculated based on the first formula, which is: Among them, x1 is the order frequency of the processing information appearing before the component fault information in the distribution vector, x2 is the order frequency of the processing information appearing after the component fault information in the distribution vector, y1 is the order frequency of the component fault information appearing before the processing information in the distribution vector, and y2 is the order frequency of the component fault information appearing after the processing information in the distribution vector.

[0087] If the distribution distance is less than or equal to the second threshold, the processing information and the component failure information are determined to be matched.

[0088] Specifically, in this embodiment, sequential occurrence frequency refers to the number of times the same phrase information appears sequentially in a reference record, where the phrase information includes component failure information and processing information. A distribution vector is a vector composed of sequential occurrence frequencies. For example, reference record P1 is sequentially "phrase information W1, phrase information J1, phrase information W1, phrase information W1, phrase information J1, phrase information J2," where "phrase information W1" represents component failure information, and "phrase information J1" and "phrase information J2" represent processing information. The first formula can be used to determine which phrase information matches the component failure information. For example, in the first formula, the frequency x1 of "phrase information J1" appearing before "phrase information W1" is 1, the frequency x2 of "phrase information J1" appearing after "phrase information W1" is 2, the frequency y1 of "phrase information W1" appearing before "phrase information J1" is 3, and the frequency y2 of "phrase information W1" appearing after "phrase information J1" is 1. Then the distribution distance D between "phrase information W1" and "phrase information J1" is Similarly, the distribution distance D between “phrase information W1” and “phrase information J2” is in, is less than the second threshold, is greater than the second threshold, so keep The corresponding matching method is that "phrase information W1" and "phrase information J1" match each other.

[0089] As a preferred technical solution of the present invention, extracting the fault analysis list includes the following steps:

[0090] Extract device fault information from the fault category based on a preset format, calculate the semantic similarity between the device fault information and the phenomenon node and the extended node in any connection formula, extract the connection formula with a semantic similarity greater than a third threshold and set it as a set of alternative connection formulas.

[0091] Based on the analysis prompt information, the connection type corresponding to the agricultural and animal husbandry equipment is extracted from the set of alternative connection types and set as a suspected fault.

[0092] Locate suspected faults in the fault knowledge base, obtain node information corresponding to the suspected fault based on the knowledge distribution table, and summarize the node information to generate a fault analysis list.

[0093] Specifically, in this embodiment, the fault category includes category information of the monitoring equipment corresponding to the abnormal data. The text recognition model can extract the equipment fault information through a preset format, and perform semantic analysis on the information corresponding to the phenomenon nodes in each connection formula to calculate the semantic similarity. The specific calculation method includes but is not limited to cosine similarity, word embedding, etc.

[0094] The analysis prompt information includes a device ID, and a connection formula corresponding to the suspected fault is screened out from multiple alternative connection formulas based on the device type corresponding to the device ID.

[0095] The connected child nodes corresponding to suspected faults can be used to locate them in the fault knowledge base. Node information refers to the nodes in the knowledge distribution table that are connected to each other and belong to suspected faults. Because the child nodes and phenomenon nodes or expansion nodes in node information contain the same information, it is necessary to aggregate all analysis nodes and combine them sequentially through directed connections to generate a fault analysis list.

[0096] like Figure 3 As shown, the present invention also provides an artificial intelligence-based equipment fault identification system, which is used to implement the above-mentioned artificial intelligence-based equipment fault identification method. The system mainly includes:

[0097] The acquisition module obtains the equipment type of any agricultural and animal husbandry equipment. The monitoring equipment collects the status monitoring data of any agricultural and animal husbandry equipment based on the pattern category, and organizes the status monitoring data corresponding to any pattern category into a historical time series based on the acquisition time. Based on the pattern category, any status monitoring data in the historical time series is marked with a fault label or a non-fault label.

[0098] The prediction module builds a fault prediction model based on all historical time series. The fault prediction model outputs the analysis sequence of agricultural and animal husbandry equipment within the prediction time, and outputs analysis prompt information based on the analysis sequence.

[0099] The identification module builds a fault knowledge base based on the equipment type, and determines whether there is a faulty device among all agricultural and animal husbandry equipment based on the analysis prompt information. If so, it outputs the fault category of the faulty device and extracts the fault analysis list corresponding to the fault category from the fault knowledge base.

[0100] The sending module sends the predicted time, fault category and fault analysis list to the management personnel.

[0101] The present invention also provides a computer storage medium storing program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the above-mentioned device fault identification method based on artificial intelligence.

[0102] It should be understood that, although the various steps in the flow chart of each embodiment of the present invention are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0103] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0104] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

[0106] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying equipment faults based on artificial intelligence, characterized in that: The method comprises the following steps: Obtaining the equipment type of any agricultural and animal husbandry equipment, using a monitoring device to collect status monitoring data of any agricultural and animal husbandry equipment based on a pattern category, organizing the status monitoring data corresponding to any pattern category into a historical time series based on a collection time, and marking any status monitoring data in the historical time series with a fault label or a non-fault label based on the pattern category; Building a fault prediction model based on all historical time series, the fault prediction model outputs an analysis sequence of the agricultural and animal husbandry equipment within a prediction time, and outputs analysis prompt information based on the analysis sequence; Among them, building a fault prediction model includes the following steps: Constructing a hybrid prediction model based on a neural network, combining all historical time series based on the acquisition time to generate a historical combination sequence, dividing the historical combination sequence into multiple subsequence combinations based on a clustering algorithm, sequentially calculating the slopes of the subsequence combinations between adjacent acquisition times, combining all the slopes and setting them as feature quantities corresponding to the subsequence combinations; Inputting all the feature quantities into the hybrid prediction model, calculating a feature score corresponding to the historical combination sequence based on the feature quantities, and if the feature score is greater than a first threshold, training the hybrid prediction model based on the feature quantities, and setting the trained hybrid prediction model as the fault prediction model; otherwise, adjusting the parameters of the clustering algorithm and repeating this step until the feature score is greater than the first threshold; Building a fault knowledge base based on the equipment type, determining whether there is a faulty device among all the agricultural and animal husbandry equipment based on the analysis prompt information, and if so, outputting the fault category of the faulty device, and extracting a fault analysis list corresponding to the fault category from the fault knowledge base; Among them, building a fault knowledge base includes the following steps: Collecting equipment manuals and historical fault records of any of the agricultural and animal husbandry equipment based on the same equipment type, and using the equipment manuals and historical fault records as reference records; Building a text recognition model based on natural language technology, setting component labels based on the device manual, extracting component failure information and processing information corresponding to the component labels from the reference records based on part-of-speech categories, and matching the component failure information with the processing information based on distribution distance; Setting any of the agricultural and animal husbandry equipment as a parent node, setting the component label as a child node, setting the component fault information as a phenomenon node, setting the matched processing information as an analysis node, performing semantic expansion on the component fault information to generate extended fault information, adding the extended fault information as an extension node in the phenomenon node, marking component connection relationships in the child nodes based on the equipment manual, and sequentially performing directed connections on the parent node, the child node, the phenomenon node, the extension node, the analysis node, and the component connection relationships to generate a knowledge distribution table; Combining the knowledge distribution tables corresponding to all the agricultural and animal husbandry equipment to generate the fault knowledge base, and performing the directed connection between the phenomenon node, the extension node, and the child node based on a preset format to generate a connection formula corresponding to the knowledge distribution table; The predicted time, the fault category and the fault analysis list are sent to a management personnel.

2. The method according to claim 1, characterized in that Calculating the feature score based on the feature quantity includes: Obtaining respectively the distribution ratios of the fault labels and the non-fault labels corresponding to all the feature quantities in the subsequence combination, and the identification ratios of the fault labels and the non-fault labels corresponding to all the feature quantities identified by the hybrid prediction model, and setting the product of the distribution ratio and the identification ratio corresponding to the same label as a first value; The largest first value is extracted from any of the subsequence combinations and set as the second value, and all the second values ​​included in the historical combination sequence are accumulated and set as the feature score.

3. The method according to claim 1, characterized in that The analysis sequence and the analysis prompt information are output based on the following steps: The hybrid prediction model includes a deep neural network and a long short-term memory network, wherein the feature quantity corresponding to any of the subsequence combinations is modeled using a mixed Gaussian distribution in an intermediate layer of the deep neural network, and a state estimation corresponding to any of the subsequence combinations is output based on the pattern category; The long short-term memory network outputs a change sequence of the historical combination sequence within the prediction time based on all the feature quantities, obtains a distribution probability of the change sequence in all the state estimates based on the mixed Gaussian distribution, and extracts the state estimate corresponding to the change sequence based on the distribution probability and sets it as the state type; The change sequence is set as the analysis sequence, a device ID is marked for the analysis sequence based on the device type, and the device ID corresponding to the analysis sequence and the state type are combined to generate the analysis prompt information.

4. The method according to claim 3, characterized in that Multiple threshold intervals are set based on the mode category and the monitoring equipment. If there is sequence data in the analysis sequence that does not belong to the threshold interval, the agricultural and animal husbandry equipment corresponding to the equipment ID is set as a faulty equipment, and the fault category corresponding to the sequence data is obtained based on the status type.

5. The method according to claim 1, wherein The matching of the component fault information with the processing information based on the distribution distance comprises the following steps: In the reference record, the sequential occurrence frequencies of any component failure information and any processing information are obtained based on a time relationship, the sequential occurrence frequencies are combined into multiple distribution vectors, and the distribution distance D is calculated based on a first formula, which is: , wherein x1 is the order occurrence frequency of the processing information appearing before the component fault information in the distribution vector, x2 is the order occurrence frequency of the processing information appearing after the component fault information in the distribution vector, y1 is the order occurrence frequency of the component fault information appearing before the processing information in the distribution vector, and y2 is the order occurrence frequency of the component fault information appearing after the processing information in the distribution vector; If the distribution distance is less than or equal to a second threshold, the processing information and the component failure information are determined to be matched.

6. The method according to claim 1, characterized in that Extracting the fault analysis list includes the following steps: Extracting device fault information from the fault category based on the preset format, calculating semantic similarity between the device fault information and the phenomenon node and the extension node in any of the connection formulas, extracting the connection formulas with semantic similarity greater than a third threshold and setting them as a set of candidate connection formulas; Extracting the connection type corresponding to the agricultural and animal husbandry equipment from the set of candidate connection types based on the analysis prompt information and setting it as a suspected fault; The suspected fault is located in the fault knowledge base, node information corresponding to the suspected fault is obtained based on the knowledge distribution table, and the node information is summarized to generate the fault analysis list.

7. An artificial intelligence-based equipment fault identification system, used to implement the method according to any one of claims 1 to 6, characterized in that: The system includes the following modules: An acquisition module obtains the equipment type of any agricultural and animal husbandry equipment, and a monitoring device collects status monitoring data of any agricultural and animal husbandry equipment based on a pattern category, organizes the status monitoring data corresponding to any pattern category into a historical time series based on the collection time, and labels any status monitoring data in the historical time series with a fault label or a non-fault label based on the pattern category; A prediction module constructs a fault prediction model based on all historical time series, the fault prediction model outputs an analysis sequence of the agricultural and animal husbandry equipment within the prediction time, and outputs analysis prompt information based on the analysis sequence; an identification module, which builds a fault knowledge base based on the equipment type, determines whether there is a faulty device among all the agricultural and animal husbandry equipment based on the analysis prompt information, and if so, outputs the fault category of the faulty device and extracts a fault analysis list corresponding to the fault category from the fault knowledge base; A sending module sends the predicted time, the fault category and the fault analysis list to a management personnel.

8. A computer storage medium, characterized in that The computer storage medium stores program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the method according to any one of claims 1 to 6.

Images