Artificial intelligence-based vector database automatic operation method and system

Abstract

The application discloses an artificial intelligence-based vector database automatic operation and maintenance method and system, relates to the technical field of automatic operation and maintenance, analyzes the equipment operation state, collects the features of the equipment with operation abnormities, analyzes the similarity degree between the current equipment operation state and the historical fault equipment operation state, stores the similar historical fault records into a temporary comparison library, judges the fault type according to the maintenance records, further obtains the predicted fault type and the number of predicted fault type records, analyzes the predicted fault type, analyzes the predicted fault type through the fault records and the current features, sorts the predicted fault type, sequentially checks the fault type after the sequence of the predicted fault type, updates the database after the fault type checking is completed, and the application reduces the complexity and uncertainty of manual operation through matching historical fault records, and shortens the fault checking time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automation operation and maintenance, in particular to a vector database automation operation and maintenance method and system based on artificial intelligence. BACKGROUND

[0002] Traditional equipment operation and maintenance highly depends on manual inspection, and its effect is limited by personnel experience and energy. In the inspection process, abnormality is easily missed due to visual fatigue and distraction of attention, especially in complex equipment systems, it is difficult to achieve comprehensive coverage. At the same time, manual inspection is limited by meteorological conditions and environmental risks, and is often forced to be interrupted in adverse weather or high-risk scenarios, resulting in potential hidden dangers that cannot be discovered in time. In addition, manual recording of data is prone to omissions and cannot be synchronized in real time, forming an information island, which makes the equipment abnormality early warning lag, and small faults are prone to develop into big problems, threatening the continuous and stable operation of the equipment; the existing fault handling relies on manual comparison of historical records by operation and maintenance personnel, and lacks systematic feature analysis and intelligent matching mechanism. In the troubleshooting process, operation and maintenance personnel need to exhaust possible paths based on experience, which is easy to be trapped in a bottleneck due to limited diagnosis direction, resulting in long and tedious troubleshooting. At the same time, manual judgment has subjective differences, which is easy to cause misjudgment or invalid maintenance attempts, causing waste of human and material resources. Especially in complex equipment fault scenarios, the traditional way is difficult to quickly lock the root cause, and the equipment downtime is often prolonged due to repeated testing, significantly increasing the operation and maintenance cost; the traditional operation and maintenance scheme is designed for specific equipment, and lacks universal adaptability, which is difficult to adapt to the operation characteristics and fault modes of different types of equipment. At the same time, fault records and maintenance data are scattered in paper documents or isolated systems, and there is no unified database support. This data island phenomenon causes the historical experience to be unable to be effectively deposited, and the new fault handling is difficult to learn from past cases, and the operation and maintenance model lacks iterative optimization basis. In the long run, the existing scheme cannot adapt to the dynamic changes of equipment operation state, and it is difficult to continuously improve the operation and maintenance capability. SUMMARY

[0003] The purpose of the present application is to provide a vector database automation operation and maintenance method and system based on artificial intelligence to solve the problems raised in the background.

[0004] In order to solve the above technical problems, the present application provides the following technical scheme: a vector database automation operation and maintenance method based on artificial intelligence, comprising the following steps:

[0005] S1, real-time monitoring of equipment operation data, analysis of equipment operation state, feature collection of equipment with operation abnormality;

[0006] S2, calling historical fault records and feature collection of equipment with operation abnormality, analyzing the similarity degree of current equipment operation state and historical fault equipment operation state;

[0007] S3, store similar historical failure records into a temporary comparison library, determine the failure type according to the maintenance record, and further obtain the predicted failure type and the number of predicted failure type records;

[0008] S4, call the device factory data, analyze the predicted failure type, and analyze the predicted failure type through the failure record and the current feature;

[0009] S5, sort the predicted failure type, and after sorting, sequentially check the failure type of the predicted failure type sequence;

[0010] S6, after completing the failure type checking, update the database.

[0011] Further, in step S1, for any type of device, collect device running data, obtain device running features after preprocessing, monitor the current device running state, and in a monitoring period, the product output by the device is α, if α / β<γ, it is judged that the current device running state is abnormal, wherein β is the standard product output per unit monitoring period, and γ is the allowed fluctuation percentage of the actual product quantity to the standard product quantity; otherwise, it is judged that the current device running state is normal, when it is judged that the current device running state is abnormal, the device running feature is called as a device running abnormal vector and stored in a running abnormal vector database, and the device label of the device to be processed is labeled, wherein the dimension of the device running abnormal vector is the same as the number of device running features; by collecting device running data and preprocessing to extract running features, and combining the allowed fluctuation range of the actual and standard product quantity to judge the state, compared with the traditional fuzzy judgment relying on experience, the accuracy of device running state judgment is improved, and the misjudgment or omission situation is reduced. When the device is abnormal, the running feature can be stored as an abnormal vector in the database and labeled as a device to be processed, the key data at the time of abnormality is completely retained, information loss is avoided, and an accurate basis is provided for subsequent fault matching and analysis. Real-time monitoring and timely abnormality identification can quickly include abnormal devices in the processing flow, avoid small abnormalities from delaying into large faults, effectively ensure the continuous and stable operation of the device, and reduce the loss caused by fault expansion.

[0012] Further, in step S2, when the device label to be processed appears in the running abnormal vector database, the device running feature corresponding to the device label to be processed is called as the current running feature, the current running feature is {A1, A2, …, AN}, wherein N represents the number of device running features, A n represents the nth current running feature, the device failure is monitored, the historical failure record is called, the number of historical failure records is M, and the device running feature corresponding to the mth failure record is {B1, B2, …, BM}, wherein M represents the number of historical failure records, and B N represents the mth historical failure record. n n N ​​}, B n represents the nth equipment running feature in the mth failure record, and then the nth feature similarity Z between the current equipment and the nth equipment of the mth failure record is obtained by considering the nth current running feature and the nth equipment running feature m_n , the nth feature dissimilarity Z of the mth failure record m_n = | (A n -B n ) / B n |, n = 1, 2, …, N, to obtain the N feature dissimilarities between the current equipment and the equipment of the mth failure record, and then the equipment dissimilarity C between the current equipment and the equipment of the mth failure record is obtained m , C m is the average of the N equipment dissimilarities between the current equipment and the equipment of the mth failure record, if C m >C0, it is judged that the equipment running features of the current equipment and the mth failure record are not similar, and C0 represents a preset equipment dissimilarity judgment threshold; otherwise, it is judged that the equipment running features of the current equipment and the mth failure record are similar; by comparing the running features of the current abnormal equipment with the features of the historical failure records, the similarity is judged based on the feature dissimilarity and the average dissimilarity, and combined with the preset threshold, the subjectivity and one-sidedness of the traditional dependence on manual experience comparison are effectively avoided, and the objectivity and accuracy of the similarity judgment are improved. This way of comprehensively considering the differences of each running feature can accurately select similar failure cases from a large number of historical records, provide reliable reference for subsequent failure type judgment, reduce blind troubleshooting, significantly improve the efficiency and pertinence of failure analysis, and lay a foundation for quickly locking the failure direction.

[0013] Further, in step S3, when it is judged that the equipment running features of the current equipment and the mth failure record are similar, the mth failure record is recorded in the temporary comparison database of the current equipment, m = 1, 2, …, M, and then X failure records are recorded in the temporary comparison database of the current equipment, and the X failure records are called as the comparison failure records of the current equipment, and then the maintenance and operation records of the X failure records are called, if the maintenance and operation records of the X failure records belong to the same failure type V after monitoring and maintenance, it is judged that the failure type of the current equipment is failure V; if the maintenance and operation records of the X failure records do not belong to the same failure type, the failure types corresponding to the X failure records are called as the predicted failure types, and the predicted failure types are {D1, D2, …, D y ,…,D Y}, where Y represents the number of predicted failure types, and D yThe number of fault records corresponding to the predicted fault types is obtained by retrieving the temporary comparison database. The number of fault records corresponding to the Y predicted fault types in the temporary comparison database is {d1,d2,…,dy,…,dY}, where dy is the predicted fault type D. y The temporary comparison database contains the number of corresponding fault records. By centrally storing historical fault records with similar characteristics to the current equipment in the temporary comparison database, and automatically analyzing fault types based on maintenance records, the fault diagnosis process is effectively simplified. When similar records belong to the same fault type, the fault type of the current equipment can be directly determined, avoiding the tedious manual comparison. When the types are different, the predicted fault type and the corresponding number of records are automatically extracted, providing a clear direction for subsequent troubleshooting. This approach reduces the subjective bias of manual judgment, ensures a more systematic and comprehensive fault type analysis, and lays a data foundation for accurate troubleshooting, improving the efficiency and reliability of fault diagnosis.

[0014] Furthermore, in step S4, the normal operating characteristics of the device at the time of factory shipment are invoked, denoted as {E1, E2, ..., E n ,…,E N}, where E n This represents the nth normal operation characteristic. For the yth predicted fault type, analysis is performed by calling the dy-th fault records of the y-th predicted fault type from the temporary comparison database. The operating characteristic of the nth device among the dy-th fault records is then obtained as {F}. 1_n ,F 2_n ,…,F w_n ,…,F dy_n}, where F w_n Let be the operating feature of the nth device in the wth fault record, where w = 1, 2, ..., dy. Then, the similarity J between the operating feature of the nth device in the wth fault record and the operating feature of the nth normal operation record can be obtained. w_n =1-|(F w_n -E n ) / E n |, if J w_n If J0 >, determine that the operating characteristics of the nth device in the wth fault record are different from the operating characteristics of the nth normal operation record; otherwise, determine that the operating characteristics of the nth device in the wth fault record are similar to the operating characteristics of the nth normal operation record.

[0015] Substitute w = 1, 2, …, dy, get the similarity of the n-th equipment running feature in dy fault records and the n-th normal running feature, and then get the number L of the n-th equipment running feature in dy fault records and the n-th normal running feature that meet the difference, if L / dy >= k0, it is judged that the n-th equipment running feature is the identification feature of dy fault records, k0 is the identification feature judgment threshold; otherwise, it is judged that the n-th equipment running feature is not the identification feature of dy fault records, and then R identification features of dy fault records are obtained, if R = 0, it is judged that dy fault records have no identification feature, and the predicted fault type D y is marked as a random fault type; otherwise, call R current running features corresponding to the identification features of dy fault records in the current running features, and the R judgment features are marked as {a1, a2, …, a r , …, a R}, where a r represents the r-th judgment feature, and the similarity of the R judgment features to the normal running features is obtained {j1, j2, …, j r , …, j R}, where j r is the similarity of the r-th judgment feature to the normal running feature, j r = 1-|(a r -E r ) / E r |, E r is the normal running feature corresponding to the r-th judgment feature in the normal running feature set, if j r > J0, it is judged that the r-th judgment feature conforms to the y-th predicted fault type; otherwise, it is judged that the r-th judgment feature does not conform to the y-th predicted fault type, and then the fault judgment priority of the y-th predicted fault type is obtained, and the fault judgment priority of the y-th predicted fault type is the ratio of the number of features conforming to the y-th predicted fault type to the number of judgment features; taking the normal running feature of the equipment out of the factory as the benchmark, the identification feature is accurately screened out by comparing and analyzing the historical records of the predicted fault type, which avoids the interference of irrelevant features on fault judgment. By judging the conformity degree of the current running feature and the identification feature, the fault priority is determined, which makes the fault analysis from fuzzy to accurate. This method not only relies on the factory standard to ensure the uniformity of the judgment benchmark, but also focuses on the key difference points through the identification feature, reduces the invalid analysis, provides clear priority basis for subsequent fault troubleshooting, improves the scientificity and pertinence of fault type judgment, and effectively avoids the problem of direction confusion caused by complex features in traditional troubleshooting.

[0016] Further, in step S5, the predicted fault types satisfying R not equal to 0 are sorted in descending order of fault judgment priority, and the random fault types satisfying R equal to 0 are sorted after the predicted fault types satisfying R not equal to 0, to obtain a predicted fault type sequence, and then the predicted fault type sequence is sequentially sorted and investigated.

[0017] Further, in step S6, after the sequential fault type investigation of the fault type sequence is completed, the temporary comparison database is deleted, and the fault information of this time is updated to the database.

[0018] The vector database automatic operation and maintenance system based on artificial intelligence comprises a data monitoring and abnormality collection module, a historical fault similarity analysis module, a similar record and type preliminary determination module, a data depth analysis module, a fault type sorting and investigation module, and an investigation result data updating module.

[0019] The data monitoring and abnormality collection module is used for monitoring equipment operation data in real time, analyzing equipment operation states, and collecting features of equipment with operation abnormalities.

[0020] The historical fault similarity analysis module is used for calling historical fault records and collecting features of equipment with operation abnormalities, and analyzing the similarity between the current equipment operation state and the historical fault equipment operation state.

[0021] The similar record and type preliminary determination module is used for storing similar historical fault records in a temporary comparison database, determining fault types according to maintenance records, and then obtaining predicted fault types and the number of predicted fault type records.

[0022] The data depth analysis module is used for calling equipment factory data, analyzing predicted fault types, and analyzing the predicted fault types through fault records and current features.

[0023] The fault type sorting and investigation module is used for sorting predicted fault types, and sequentially sorting and investigating the predicted fault type sequence after the sorting.

[0024] The investigation result data updating module is used for updating the database after the fault type investigation is completed.

[0025] Compared with the prior art, the present application has the beneficial effects that: on the one hand, by collecting equipment operation data in real time and preprocessing to extract features, combined with preset standards to accurately judge the operation state, abnormal equipment can be quickly identified and abnormal features are retained. Compared with the traditional manual inspection method, the delay in finding abnormalities caused by human oversight can be avoided, abnormal equipment can be timely included in the processing flow, small faults can be prevented from developing into big problems, continuous and stable operation of equipment can be effectively ensured, production interruption and increased maintenance costs caused by equipment downtime or fault expansion can be reduced, and active early warning support for reliable operation of equipment is provided.

[0026] On the one hand, in the fault handling link, the method matches historical fault records, analyzes feature similarity, filters similar cases, and then combines maintenance records and factory data to deepen fault type analysis, and finally sequentially investigates according to priority. This process does not require manual comparison of a large number of historical records or blind testing, reduces the complexity and uncertainty of manual operation, and shortens the time consumption of fault investigation. At the same time, according to the historical data and feature analysis, the fault direction is accurately locked, invalid maintenance attempts are avoided, human and material resources are wasted, the operation and maintenance cost is significantly reduced, and the overall operation and maintenance efficiency is improved;

[0027] On the other hand, after each fault investigation is completed, the database is updated and new fault and maintenance data are deposited, providing more abundant reference basis for subsequent operation and maintenance. With the accumulation of data, the historical fault matching is more accurate, the fault type judgment is more in line with the actual equipment operation condition, and the operation and maintenance model is gradually optimized. In addition, the method can adapt to the operation characteristics and fault modes of different types of equipment, without the need to design an operation and maintenance scheme for a single equipment, and has strong universality. In the long run, data closed loop and universal adaptability make the system continuously adapt to the change of equipment operation state, continuously improve the operation and maintenance reliability and intelligent level, and provide sustainable support for long-term stable operation and maintenance of equipment. BRIEF DESCRIPTION OF DRAWINGS

[0028] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0029] Figure 1 is a structure diagram of the vector database automatic operation and maintenance system based on artificial intelligence of the present application;

[0030] Figure 2 is a flowchart of the vector database automatic operation and maintenance method based on artificial intelligence of the present application. DETAILED DESCRIPTION

[0031] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0032] Please refer to Figure 1 and Figure 2 The present application provides a technical solution: an artificial intelligence-based vector database automatic operation and maintenance method, comprising the following steps:

[0033] S1, real-time monitoring of equipment operation data, analysis of equipment operation state, feature collection of equipment with operation anomaly;

[0034] S2, calling historical fault records and feature collection of equipment with operation anomaly, analyzing the similarity between the current equipment operation state and the historical fault equipment operation state;

[0035] S3, storing similar historical fault records into a temporary comparison library, judging the fault type according to the repair record, and then obtaining the predicted fault type and the number of predicted fault type records;

[0036] S4, calling equipment factory data, analyzing the predicted fault type, and analyzing the predicted fault type through the fault record and the current feature;

[0037] S5, sorting the predicted fault type, and sequentially troubleshooting the predicted fault type sequence after sorting;

[0038] S6, after completing the fault type troubleshooting, updating the database.

[0039] In step S1, for any type of equipment, equipment operation data is collected, and after preprocessing, equipment operation characteristics are obtained. The current equipment operation status is monitored. In one monitoring cycle, the product output of the equipment is α. If α / β < γ, the current equipment operation status is judged to be abnormal, where β is the predetermined standard product output per unit monitoring cycle, and γ is the predetermined allowable fluctuation percentage of the ratio of actual product output to standard product output. Otherwise, the current equipment operation status is judged to be normal. When the current equipment operation status is judged to be abnormal, the equipment operation characteristics are called as equipment operation anomaly vectors and stored in the operation anomaly vector database. The equipment tag of the equipment to be processed is marked, where the dimension of the equipment operation anomaly vector is the same as the number of equipment operation characteristics. By collecting equipment operation data and preprocessing to extract operation characteristics, and combining the allowable fluctuation range of the ratio of actual to standard product output to determine the status, compared with the traditional fuzzy judgment relying on experience, the accuracy of equipment operation status judgment is improved, and the situation of misjudgment or omission is reduced. When the equipment is abnormal, the operation characteristics can be stored as anomaly vectors in the database and marked with tags to be processed, and the key data at the time of the anomaly can be completely retained to avoid information loss and provide an accurate basis for subsequent fault matching and analysis. Real-time monitoring and timely anomaly identification can quickly incorporate abnormal equipment into the processing flow, preventing minor anomalies from escalating into major failures, effectively ensuring the continuous and stable operation of equipment, and reducing losses caused by the expansion of failures.

[0040] In step S2, when a device tag to be processed appears in the anomaly vector database, the device operation feature corresponding to the device tag to be processed is called as the current operation feature. The current operation feature is {A1, A2, ..., A...} n ,…,A N}, where N represents the number of equipment operating characteristics, A n This represents the nth current operating characteristic. It monitors equipment faults and retrieves historical fault records. The number of historical fault records is M, and the equipment operating characteristic corresponding to the mth fault record is {B1, B2, ..., B...}. n ,…,B N}, B n This represents the operating feature of the nth device in the mth fault record. Then, considering the nth current operating feature and the operating feature of the nth device, we obtain the similarity Z between the nth feature of the current device and the device in the mth fault record. m_n The dissimilarity Z of the nth feature of the mth fault record m_n =|(A n -B n ) / B n Substitute each of the following into n=1,2,…,N to obtain the N feature dissimilarity between the current device and the device in the m-th fault record, and then obtain the device dissimilarity C between the current device and the device in the m-th fault record. m C mis the average of the N device dissimilarity of the current device and the mth fault record device, if C m > C0, it is judged that the device running characteristics of the current device are not similar to the device running characteristics of the mth fault record device, C0 represents a preset device dissimilarity judgment threshold; otherwise, it is judged that the device running characteristics of the current device are similar to the device running characteristics of the mth fault record device; the running characteristics of the current abnormal device are compared with the characteristics of the historical fault records through a system call, similarity is judged based on the feature dissimilarity and the average dissimilarity, and in combination with a preset threshold, subjectivity and one-sidedness of traditional dependence on manual experience comparison are effectively avoided, and objectivity and accuracy of similarity judgment are improved. This way of comprehensively considering the differences of each running characteristic can accurately filter out similar fault cases from a large number of historical records, provide reliable reference for subsequent fault type judgment, reduce blind troubleshooting, significantly improve the efficiency and pertinence of fault analysis, and lay a foundation for quickly locking the fault direction.

[0041] In step S3, when it is judged that the device running characteristics of the current device are similar to the device running characteristics of the mth fault record device, the mth fault record is recorded in the temporary comparison database of the current device, m = 1, 2, …, M is substituted one by one, and then X fault records are recorded in the temporary comparison database of the current device, the X fault records are called as comparison fault records of the current device, and then the maintenance and operation records of the X fault records are called, if the maintenance and operation records of the X fault records belong to the same fault type V after monitoring and maintenance, it is judged that the fault type of the current device is fault V; if the maintenance and operation records of the X fault records do not belong to the same fault type, the fault types corresponding to the X fault records are called as predicted fault types, and the predicted fault types are {D1, D2, …, D y ,…,D Y}, where Y represents the number of predicted fault types, D y is the predicted fault type, and dy is the number of fault records corresponding to the predicted fault type D y in the temporary comparison database. By storing the historical fault records similar to the current device characteristics in the temporary comparison database, the fault type is automatically analyzed relying on the maintenance and operation records, and the fault judgment process is effectively simplified. When the similar records belong to the same fault type, the fault type of the current device can be directly determined, avoiding the tediousness of manual comparison one by one; when the types are different, the predicted fault types and the corresponding record numbers are automatically extracted, providing a clear direction for subsequent troubleshooting. This way reduces the subjective bias of manual judgment, ensures that the fault type analysis is more systematic and comprehensive, lays a data foundation for accurate troubleshooting, and improves the efficiency and reliability of fault determination.

[0041] In step S3, when it is judged that the device running characteristics of the current device are similar to the device running characteristics of the mth fault record device, the mth fault record is recorded in the temporary comparison database of the current device, m = 1, 2, …, M is substituted one by one, and then X fault records are recorded in the temporary comparison database of the current device, the X fault records are called as comparison fault records of the current device, and then the maintenance and operation records of the X fault records are called, if the maintenance and operation records of the X fault records belong to the same fault type V after monitoring and maintenance, it is judged that the fault type of the current device is fault V; if the maintenance and operation records of the X fault records do not belong to the same fault type, the fault types corresponding to the X fault records are called as predicted fault types, and the predicted fault types are {D1, D2, …, D y ,…,D Y}, where Y represents the number of predicted fault types, D y is the predicted fault type, and dy is the number of fault records corresponding to the predicted fault type D y in the temporary comparison database. By storing the historical fault records similar to the current device characteristics in the temporary comparison database, the fault type is automatically analyzed relying on the maintenance and operation records, and the fault judgment process is effectively simplified. When the similar records belong to the same fault type, the fault type of the current device can be directly determined, avoiding the tediousness of manual comparison one by one; when the types are different, the predicted fault types and the corresponding record numbers are automatically extracted, providing a clear direction for subsequent troubleshooting. This way reduces the subjective bias of manual judgment, ensures that the fault type analysis is more systematic and comprehensive, lays a data foundation for accurate troubleshooting, and improves the efficiency and reliability of fault determination.

[0042] In step S4, the normal operating characteristics of the device at the time of factory shipment are invoked, denoted as {E1, E2, ..., E n ,…,E N}, where E n This represents the nth normal operation characteristic. For the yth predicted fault type, analysis is performed by calling the dy-th fault records of the y-th predicted fault type from the temporary comparison database. The operating characteristic of the nth device among the dy-th fault records is then obtained as {F}. 1_n ,F 2_n ,…,F w_n ,…,F dy_n}, where F w_n Let be the operating feature of the nth device in the wth fault record, where w = 1, 2, ..., dy. Then, the similarity J between the operating feature of the nth device in the wth fault record and the operating feature of the nth normal operation record can be obtained. w_n =1-|(F w_n -E n ) / E n |, if J w_n If J0 >, determine that the operating characteristics of the nth device in the wth fault record are different from the operating characteristics of the nth normal operation record; otherwise, determine that the operating characteristics of the nth device in the wth fault record are similar to the operating characteristics of the nth normal operation record.

[0043] Substitute each value of w = 1, 2, ..., dy into the list of fault records to obtain the similarity between the nth device's operating feature and the nth normal operation feature. Then, obtain the number L of dissimilar features between the nth device's operating feature and the nth normal operation feature in the dy fault records. If L / dy ≥ k0, then determine that the nth device's operating feature is an identifier feature of the dy fault records, where k0 is a pre-set identifier feature judgment threshold. Otherwise, determine that the nth device's operating feature is not an identifier feature of the dy fault records, and obtain R identifier features of the dy fault records. If R = 0, then determine that the dy fault records have no identifier features, and predict the fault type D. y If the fault is classified as random, then the R current operating features corresponding to the identifier features of the dy fault records are called. These R current operating features are used as judgment features, and are labeled as {a1, a2, ..., a...}. r ,…,a R}, where a r Let represent the r-th judgment feature, and obtain the similarity {j1,j2,…,j} of the R judgment features to the normal operation feature. r ,…,j R}, where j r Let j be the similarity between the r-th judgment feature and the normal operation feature. r =1-|(a r-E r ) / E r |,E r is the normal operation feature corresponding to the rth judgment feature in the normal operation feature set, if j r >J0, it is judged that the rth judgment feature conforms to the yth predicted fault type; otherwise, it is judged that the rth judgment feature does not conform to the yth predicted fault type, and the fault judgment priority of the yth predicted fault type is obtained, the fault judgment priority of the yth predicted fault type is the ratio of the number of features conforming to the yth predicted fault type to the number of judgment features; taking the normal operation feature of the equipment out of the factory as the benchmark, the identification features are accurately screened out by comparing and analyzing the historical records of the predicted fault types, and the interference of irrelevant features on fault judgment is avoided. The fault priority is determined by judging the conformity degree of the current operation feature and the identification feature, so that the fault analysis goes from fuzzy to accurate. This way not only relies on the factory standard to ensure the uniformity of the judgment benchmark, but also focuses on the key differences through identification features, reduces invalid analysis, provides clear priority basis for subsequent fault troubleshooting, improves the scientificity and pertinence of fault type judgment, and effectively avoids the problem of direction confusion caused by complex features in traditional troubleshooting.

[0044] In step S5, the predicted fault types satisfying R not equal to 0 are sorted from high to low according to the fault judgment priority, and the random fault types satisfying R equal to 0 are sorted after the predicted fault types satisfying R not equal to 0, to obtain a predicted fault type sequence, and then the predicted fault type sequence is sequentially fault type troubleshooting.

[0045] In step S6, after completing the sequential fault type troubleshooting of the fault type sequence, the temporary comparison database is deleted, and the fault information of this time is updated to the database.

[0046] The vector database automatic operation and maintenance system based on artificial intelligence, the system comprises: a data monitoring and abnormality collection module, a historical fault similarity analysis module, a similar record and type preliminary determination module, a data depth analysis module, a fault type sorting and troubleshooting module, and a troubleshooting result data updating module;

[0047] The data monitoring and abnormality collection module is used for real-time monitoring of equipment operation data, analyzing the equipment operation state, and collecting features of the equipment with operation abnormalities;

[0048] The historical fault similarity analysis module is used for calling historical fault records and collecting features of the equipment with operation abnormalities, and analyzing the similarity between the current equipment operation state and the historical fault equipment operation state;

[0049] The similar record and type initial determination module is used for storing similar historical fault records into a temporary comparison database, judging a fault type according to a maintenance record, and then obtaining a predicted fault type and a number of predicted fault type records;

[0050] The data deep analysis module is used for calling equipment factory data, analyzing a predicted fault type, and analyzing the predicted fault type through a fault record and a current feature;

[0051] The fault type sorting and troubleshooting module is used for sorting a predicted fault type, and sequentially troubleshooting the predicted fault type sequence after sorting.

[0052] The troubleshooting result data updating module is used for updating a database after completing fault type troubleshooting.

[0053] In step S1, for the processing equipment, real-time collection is performed on data such as temperature, rotating speed, energy consumption, and vibration frequency in the running process of the equipment, abnormal values of temporary fluctuations are removed through data cleaning, and then key features reflecting the running state of the equipment are obtained through feature extraction and integration. Meanwhile, the number of qualified parts produced by the equipment in a complete processing cycle is monitored, and the number is compared with a pre-set standard qualified product quantity per unit cycle. If the ratio of the actual qualified quantity to the standard quantity is lower than the allowed fluctuation range, it is determined that the equipment is running abnormally, the extracted equipment running features are stored as an abnormal vector in an abnormal vector database, and a to-be-processed label is marked for the equipment; if the ratio is within the allowed range, it is determined that the equipment is running normally, and real-time monitoring is continued.

[0054] After entering step S2, when the system detects the to-be-processed label of the equipment in the abnormal vector database, the current running features of the equipment corresponding to the label are called, and the past fault records of the same type of processing equipment stored in the database are called. The features such as temperature and rotating speed of the current equipment are compared with the corresponding features in each historical fault record one by one, the difference degree between the two is analyzed, the average value of all feature differences is calculated, and the historical fault records similar to the current equipment state are screened out in combination with a pre-set difference judgment threshold.

[0055] In step S3, the screened similar historical fault records are stored in the temporary comparison database of the equipment, and then the maintenance and operation archives corresponding to the historical records are called. If the archives show that the records all correspond to a "bearing wear" fault type, it is preliminarily determined that the fault type of the current equipment is bearing wear; if the archives contain multiple fault types such as bearing wear and motor overload, the types are taken as the predicted fault types of the current equipment, and the number of historical records corresponding to each predicted type in the temporary comparison database is counted.

[0056] In step S4, the normal operation characteristic data of the processing equipment when it leaves the factory is called, including the standard temperature range, the rated rotating speed, the normal energy consumption interval, etc., and for each predicted fault type, the difference between the equipment characteristics in the corresponding historical record and the normal characteristics when leaving the factory is analyzed. The characteristics that generally deviate from the normal range in this type of fault are selected as the identification characteristics, such as the abnormal temperature rise characteristics often appearing in bearing wear faults, and then the compliance degree of the corresponding characteristics of the current equipment with the identification characteristics is compared, and the judgment priority of each predicted fault type is calculated.

[0057] In step S5, the predicted fault types are sorted according to the judgment priority from high to low, and the random fault types without clear identification characteristics are arranged at the end to form a fault troubleshooting sequence. The operation and maintenance personnel carry out troubleshooting according to the sequence, preferentially check the bearing wear faults of high priority, disassemble the bearing part of the equipment to check the wear condition, and if no problem is found, then sequentially troubleshoot the motor overload and other types.

[0058] After completing the fault troubleshooting and repairing the equipment, enter step S6, delete the temporary comparison database of the equipment, and update the equipment fault type, troubleshooting process, repair scheme, etc. to the system database, providing data support for the operation and maintenance of subsequent similar equipment.

[0059] It is apparent to those skilled in the art that the present application is not limited to the details of the above exemplary sensor device embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be considered as limiting the claims involved.

Claims

1. An artificial intelligence-based vector database automatic operation and maintenance method, characterized in that: The method comprises the following steps: S1, real-time monitoring equipment operation data, analyzing the equipment operation state, collecting features of the equipment running abnormally; S2, calling the historical fault records and the features of the equipment running abnormally, analyzing the similarity between the current equipment operation state and the historical fault equipment operation state; S3, storing similar historical fault records in a temporary comparison database, judging the fault type according to the maintenance record, and then obtaining the predicted fault type and the number of predicted fault type records; S4, calling the equipment factory data, analyzing the predicted fault type, and analyzing the predicted fault type through the fault record and the current feature; S5, sorting the predicted fault type, and sequentially checking the fault type after sorting the predicted fault type sequence; S6, updating the database after completing the fault type checking; In step S1, for any type of equipment, the equipment operation data is collected, the equipment operation features are obtained after preprocessing, the current equipment operation state is monitored, and in a monitoring period, the product quantity of the equipment is α, if α / β<γ, it is judged that the current equipment operation state is abnormal, wherein β is the standard product quantity of the unit monitoring period, and γ is the allowed fluctuation percentage of the actual product quantity and the standard product quantity ratio; otherwise, it is judged that the current equipment operation state is normal, when it is judged that the current equipment operation state is abnormal, the equipment operation features are called as equipment operation abnormal vector and stored in the running abnormal vector database, and the equipment label of the equipment to be processed is labeled, wherein the dimension of the equipment operation abnormal vector is the same as the number of the equipment operation features; In step S2, when a device tag to be processed appears in the anomaly vector database, the device operation feature corresponding to the device tag to be processed is called as the current operation feature. The current operation feature is {A1, A2, ..., A...} n ,…,A N }, where N represents the number of equipment operating characteristics, A n This represents the nth current operating characteristic. It monitors equipment faults and retrieves historical fault records. The number of historical fault records is M, and the equipment operating characteristic corresponding to the mth fault record is {B1, B2, ..., B...}. n ,…,B N }, B n This represents the operating characteristic of the nth device in the mth fault record. Then, considering the nth current operating characteristic and the operating characteristic of the nth device, we obtain the similarity Z between the nth feature of the current device and the device in the mth fault record. m_n The dissimilarity Z of the nth feature of the mth fault record m_n =|(A n -B n ) / B n Substitute each of the following into n=1,2,…,N to obtain the N feature dissimilarity between the current device and the device in the m-th fault record, and then obtain the device dissimilarity C between the current device and the device in the m-th fault record. m C m Let C be the average of the dissimilarity of the current device and the device with the m-th fault record among the N devices. m If >C0, it is determined that the operating characteristics of the current device are not similar to those of the device in the m-th fault record. C0 represents the pre-set threshold for determining the dissimilarity of the devices. Otherwise, it is determined that the operating characteristics of the current device are similar to those of the device in the m-th fault record. 2.The artificial intelligence based vector database automation operation and maintenance method according to claim 1, characterized in that: In step S3, when it is judged that the device running feature of the current device is similar to the device running feature of the mth fault record, the mth fault record is recorded in the temporary comparison database of the current device, and m=1, 2, …, M is substituted in turn, and then X fault records are recorded in the temporary comparison database of the current device, the X fault records are called as the comparison fault records of the current device, and the maintenance and operation records of the X fault records are called, if the maintenance and operation records of the X fault records belong to the same fault type V after monitoring and maintenance, it is judged that the fault type of the current device is fault V; if the maintenance and operation records of the X fault records do not belong to the same fault type after monitoring and maintenance, the fault types corresponding to the X fault records are called as the predicted fault types, and the predicted fault types are {D1, D2, …, D y ,…,D Y}, wherein Y represents the number of predicted fault types, D y , the number of fault records corresponding to the predicted fault types is obtained by searching the temporary comparison database, and the number of fault records corresponding to the predicted fault types Y in the temporary comparison database is {d1, d2, …, dy, …, dY}, wherein dy is the number of fault records corresponding to the predicted fault type D y in the temporary comparison database.

3. The artificial intelligence-based vector database automation operation and maintenance method according to claim 2, characterized in that: In step S4, the normal operating characteristics of the device at the time of factory shipment are invoked, denoted as {E1, E2, ..., E n ,…,E N }, where E n This represents the nth normal operation characteristic. For the yth predicted fault type, analysis is performed by calling the dy-th fault records of the y-th predicted fault type from the temporary comparison database. The operating characteristic of the nth device among the dy-th fault records is then obtained as {F}. 1_n ,F 2_n ,…,F w_n ,…,F dy_n }, where F w_n Let be the operating feature of the nth device in the wth fault record, where w = 1, 2, ..., dy. Then, the similarity J between the operating feature of the nth device in the wth fault record and the operating feature of the nth normal operation record can be obtained. w_n =1-|(F w_n -E n ) / E n |, if J w_n >J0, determine that the operating characteristics of the nth device in the wth fault record are different from the operating characteristics of the nth normal operation record; Otherwise, it is judged that the nth equipment operation feature in the wth fault record is similar to the nth normal operation feature.

4. The artificial intelligence-based vector database automated operation and maintenance method according to claim 3, characterized in that: Substitute w=1,2,…,dy, get the similarity of the nth equipment operation feature in the dy fault records and the nth normal operation feature, and then get the number L of the nth equipment operation feature in the dy fault records and the nth normal operation feature satisfying the difference, if L / dy≥k0, it is judged that the nth equipment operation feature is the identification feature of the dy fault records, and k0 is the identification feature judgment threshold; Otherwise, it is judged that the n-th equipment operation characteristic is not the identification characteristic of the dy fault records, and then R identification characteristics of the dy fault records are obtained. If R=0, it is judged that the dy fault records have no identification characteristics, and the predicted fault type D y is marked as a random fault type; otherwise, R current operation characteristics corresponding to the identification characteristics of the dy fault records in the current operation characteristics are called, and the R current operation characteristics are used as judgment characteristics. The R judgment characteristics are marked as {a1, a2, …, a r ,…,a R}, wherein a r represents the r-th judgment characteristic, and the similarity of the R judgment characteristics to the normal operation characteristics is obtained as {j1, j2, …, j r ,…,j R}, wherein j r is the similarity of the r-th judgment characteristic to the normal operation characteristics, j r =1-|(a r -E r ) / E r |, E r is the normal operation characteristic corresponding to the r-th judgment characteristic in the normal operation characteristic set, and if j r >Jo, it is judged that the r-th judgment characteristic conforms to the y-th predicted fault type; otherwise, it is judged that the r-th judgment characteristic does not conform to the y-th predicted fault type, and then the fault judgment priority of the y-th predicted fault type is obtained. The fault judgment priority of the y-th predicted fault type is the ratio of the number of characteristics conforming to the y-th predicted fault type to the number of judgment characteristics. 5.The artificial intelligence based vector database automation operation and maintenance method of claim 4, characterized in that: In step S5, the predicted fault types satisfying R not equal to 0 are sorted from high to low according to the fault judgment priority, the random fault types satisfying R equal to 0 are sorted after the predicted fault types satisfying R not equal to 0, and the predicted fault type sequence is obtained, and then the fault type sequence is sequentially checked. 6.The artificial intelligence based vector database automation operation and maintenance method according to claim 5, characterized in that: In step S6, after completing the sequential fault type checking of the fault type sequence, the temporary comparison database is deleted, and the fault information is updated in the database.

7. The system for automatically operating and maintaining the vector database based on artificial intelligence, applied to the method for automatically operating and maintaining the vector database based on artificial intelligence in any one of claims 1-6, characterized in that: The system comprises: a data monitoring and abnormality collection module, a historical fault similarity analysis module, a similar record and type preliminary determination module, a data depth analysis module, a fault type sorting and checking module, and a checking result data updating module; The data monitoring and abnormality collection module is used for real-time monitoring equipment operation data, analyzing the equipment operation state, and collecting features of the equipment running abnormally; The historical fault similarity analysis module is configured to call historical fault records and collect features of the running abnormal equipment, and analyze the similarity between the current equipment running state and the historical fault equipment running state; The similar record and type preliminary determination module is configured to store similar historical fault records into a temporary comparison library, determine the fault type according to the maintenance record, and further obtain the predicted fault type and the number of predicted fault type records; The data depth analysis module is configured to call the equipment factory data, analyze the predicted fault type, and analyze the predicted fault type through the fault record and the current feature; The fault type sorting and troubleshooting module is configured to sort the predicted fault type, and sequentially troubleshoot the predicted fault type sequence after sorting; The troubleshooting result data updating module is configured to update the database after completing the fault type troubleshooting.

Images