Alarm management method and device based on AIoT

Abstract

The invention discloses an AIoT-based alarm management method and device, and the method comprises the steps: firstly collecting multi-dimensional basic data through an AIoT terminal, enabling a device management system and a user operation platform to be connected to obtain a device historical record and a user operation track, uploading the historical record and the user operation track to a cloud platform after preprocessing, and constructing a multi-dimensional data set; identifying valid and invalid alarm data from the data set by using a random forest and isolated forest combined algorithm; combining the data set and alarm data, analyzing historical fault, equipment aging and environment trend characteristics through a bidirectional LSTM neural network, outputting a fault risk value, and triggering a pre-alarm; then, constructing an equipment association graph by means of a graph neural network, positioning multi-equipment alarm sources, and merging alarms; and finally, the AIoT visual collaborative operation and maintenance platform dispatches an order according to a rule to track the progress, a statistical module generates a report, and alarm full-life-cycle management is realized. According to the invention, the problems of high false alarm rate, lack of fault prediction capability and the like in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things (IoT) and alarm management technology, specifically to an alarm management method and device based on AIoT. Background Technology

[0002] With the deep integration of Internet of Things (IoT) technology with industrial production and equipment management scenarios, AIoT has become a core technology direction for improving equipment operation and maintenance efficiency, and various equipment alarm management systems have emerged. Among the existing technologies, the "Method and System for Processing IoT Device Alarm Messages" disclosed in CN113518006A is representative. Its core application logic is as follows: after receiving alarm messages reported by IoT devices, a group identifier is constructed according to "device serial number and alarm type", a group queue and a group record table are matched or created, the duplication of alarm messages with historical messages in the group is determined by MD5 checksum, and the data collection cycle is controlled by combining the sampling time window threshold. When the duplication counter reaches the preset value, key alarm information is filtered and stored, thereby reducing the amount of duplicate alarm processing and reducing platform resource consumption. This type of solution has been applied to some extent in the basic alarm management of small and medium-sized IoT device clusters.

[0003] However, existing technologies still have significant limitations in complex real-world scenarios, making it difficult to meet the refined operation and maintenance needs of enterprises: First, the data dimensions are limited, only filtering alarm messages based on their repetitiveness, without integrating equipment operation data, environmental data, network data, and user operation logs. This makes it impossible to distinguish between "invalid duplicate alarms" and "valid similar alarms," ​​resulting in a false alarm rate as high as 25%, wasting a significant amount of operation and maintenance effort. Second, there is a lack of predictive alarm capabilities, passively filtering and storing only reported alarm messages. It is impossible to predict potential faults in advance based on equipment historical fault records, aging parameters, and environmental trends. The average time from alarm reporting to fault resolution is 1.5 hours, which can easily cause significant production interruption losses. Third, the functional coverage is incomplete. A collaborative operation and maintenance mechanism has not been built. It only completes the filtering and storage of alarm information, without connecting to the operation and maintenance personnel resource pool. It lacks an automatic work order allocation function based on "skill matching degree and location distance," and there is no alarm processing progress tracking and historical data statistics module. The work order response delay rate reaches 30%, making it impossible to achieve full lifecycle management of alarms.

[0004] Therefore, there is an urgent need for an AIoT-based alarm management method to solve the problems of high false alarm rate, lack of fault prediction capability based on multi-source data, and inability to cover the entire lifecycle management of alarms without a collaborative operation and maintenance mechanism. Summary of the Invention

[0005] To address these issues, the present invention provides an alarm management method and apparatus based on AIoT, which solves the problems of existing IoT device alarm management systems, such as high false alarm rates due to single data dimensions, lack of fault prediction capabilities based on multi-source data, and inability to cover the entire lifecycle management of alarms due to the lack of collaborative operation and maintenance mechanisms.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an alarm management method based on AIoT, characterized in that it includes:

[0007] Multi-dimensional basic data is collected through AIoT terminals; device history records and user operation trajectories are obtained by connecting to the device management system and user operation platform; the multi-dimensional basic data, device history records and user operation trajectories are preprocessed to obtain preprocessed data and uploaded to the cloud platform; a multi-dimensional dataset is constructed based on the preprocessed data.

[0008] By combining random forest and isolated forest algorithms, the multi-dimensional features in the multi-dimensional dataset are classified and calculated to identify valid alarm data and invalid alarm data.

[0009] Based on the multi-dimensional dataset, combined with the valid alarm data and the invalid alarm data, a bidirectional LSTM neural network is used to analyze historical fault characteristics, equipment aging characteristics and environmental trend characteristics, output a fault risk value for a specified period in the future, and trigger a corresponding level of pre-alarm based on the fault risk value.

[0010] Based on the multi-dimensional dataset, combined with the valid alarm data and pre-alarm data, a device association graph is constructed through a graph neural network. When several devices generate valid alarms or pre-alarms at the same time, the abnormal state of the associated nodes is analyzed through the device association graph to locate the root cause of the fault and merge duplicate alarms to generate root cause alarm information.

[0011] Based on the root cause alarm information and pre-alarm level, work orders are allocated and processing progress is tracked through the AIoT visual collaborative operation and maintenance platform according to the set allocation principles; through the statistical analysis module, operation and maintenance efficiency reports and alarm trend reports are generated based on alarm data and work order processing data to realize alarm lifecycle management.

[0012] As a preferred solution for an AIoT-based alarm management method, the multi-dimensional basic data includes: motor vibration, temperature, voltage / current collected by device sensors; temperature, humidity, dust concentration, and air pressure monitored by environmental sensors; and bandwidth, latency, and packet loss rate obtained by the network monitoring module.

[0013] The equipment history record includes equipment maintenance records and fault logs for the past 90 days.

[0014] The user operation trajectory is a record of abnormal device start-ups and shutdowns and parameter modifications over the past 24 hours.

[0015] During the preprocessing of the multi-dimensional basic data, the device historical records, and the user operation trajectory, outlier removal and duplicate data deletion are performed through edge computing nodes to obtain the preprocessed data.

[0016] As a preferred solution for an AIoT-based alarm management method, in the process of classifying and calculating the multi-dimensional features in the multi-dimensional dataset, the multi-dimensional features include: device features, environmental features, network features, and user features.

[0017] The equipment features include real-time vibration frequency, temperature, voltage, cumulative operating time, and most recent maintenance time;

[0018] The environmental characteristics are the temperature and humidity fluctuation values ​​and the peak dust concentration over the past hour.

[0019] The network characteristics are the average latency and packet loss rate over the past 30 minutes;

[0020] The user characteristics are the number of abnormal operations in the past 24 hours.

[0021] As a preferred solution for an AIoT-based alarm management method, in the process of analyzing the historical fault characteristics, the equipment aging characteristics, and the environmental trend characteristics through the bidirectional LSTM neural network, the historical fault characteristics are the occurrence time of similar equipment failures and the trend of parameter changes before failures in the past 90 days; the equipment aging characteristics are the ratio of cumulative running time to design life and the replacement cycle of key components; and the environmental trend characteristics are the slope of temperature and humidity changes in the past 7 days and environmental prediction data for the next 24 hours.

[0022] As a preferred solution for an AIoT-based alarm management method, the device association graph nodes include: devices, power supply lines, environmental areas, and network switches, with edges representing power supply association, network association, and location association.

[0023] As a preferred solution for alarm management based on AIoT, in the process of allocating work orders and tracking processing progress through the AIoT visual collaborative operation and maintenance platform according to the set allocation principles, work orders are allocated based on the principle of combining skill matching degree and proximity, combined with the skill tags of operation and maintenance personnel, the distance of the fault point and the current number of work orders.

[0024] The present invention also provides an AIoT-based alarm management device, which, based on the above-mentioned AIoT-based alarm management method, includes:

[0025] A multi-dimensional dataset construction unit is used to collect multi-dimensional basic data through AIoT terminals; obtain device history records and user operation trajectories by connecting to the device management system and the user operation platform; preprocess the multi-dimensional basic data, the device history records and the user operation trajectories to obtain preprocessed data and upload it to the cloud platform; and construct a multi-dimensional dataset based on the preprocessed data.

[0026] The valid alarm data and invalid alarm data identification unit is used to classify and calculate the multi-dimensional features in the multi-dimensional dataset using a combination algorithm of random forest and isolated forest, and to identify valid alarm data and invalid alarm data.

[0027] The early warning unit is used to analyze historical fault characteristics, equipment aging characteristics and environmental trend characteristics based on the multi-dimensional dataset, combined with the valid alarm data and the invalid alarm data, through a bidirectional LSTM neural network, output a fault risk value for a specified period in the future, and trigger a corresponding level of early warning based on the fault risk value.

[0028] The root cause alarm information generation unit is used to construct a device association graph through a graph neural network based on the multi-dimensional dataset, combined with the valid alarm data and the pre-alarm data; when several devices generate valid alarms or pre-alarms at the same time, the abnormal state of the associated nodes is analyzed through the device association graph to locate the root cause of the fault and merge duplicate alarms to generate root cause alarm information.

[0029] The operation and maintenance efficiency report and alarm trend report generation unit is used to allocate work orders and track processing progress through the AIoT visual collaborative operation and maintenance platform according to the set allocation principles based on the root cause alarm information and pre-alarm level; and to generate operation and maintenance efficiency reports and alarm trend reports based on alarm data and work order processing data through the statistical analysis module, so as to realize alarm full life cycle management.

[0030] As a preferred solution for an AIoT-based alarm management device, the multi-dimensional dataset construction unit includes the following multi-dimensional basic data: motor vibration, temperature, and voltage / current collected by device sensors; temperature, humidity, dust concentration, and air pressure monitored by environmental sensors; and bandwidth, latency, and packet loss rate obtained by the network monitoring module.

[0031] The equipment history record includes equipment maintenance records and fault logs for the past 90 days.

[0032] The user operation trajectory is a record of abnormal device start-ups and shutdowns and parameter modifications over the past 24 hours.

[0033] During the preprocessing of the multi-dimensional basic data, the device historical records, and the user operation trajectory, outlier removal and duplicate data deletion are performed through edge computing nodes to obtain the preprocessed data.

[0034] As a preferred solution for an AIoT-based alarm management device, in the valid alarm data and invalid alarm data identification unit, during the classification calculation of the multi-dimensional features in the multi-dimensional dataset, the multi-dimensional features include: device features, environmental features, network features, and user features.

[0035] The equipment features include real-time vibration frequency, temperature, voltage, cumulative operating time, and most recent maintenance time;

[0036] The environmental characteristics are the temperature and humidity fluctuation values ​​and the peak dust concentration over the past hour.

[0037] The network characteristics are the average latency and packet loss rate over the past 30 minutes;

[0038] The user characteristics are the number of abnormal operations in the past 24 hours.

[0039] As a preferred solution for an AIoT-based alarm management device, the pre-alarm unit analyzes the historical fault characteristics, equipment aging characteristics, and environmental trend characteristics using the bidirectional LSTM neural network. The historical fault characteristics are the occurrence time of similar equipment failures and the trend of parameter changes before failures in the past 90 days; the equipment aging characteristics are the ratio of cumulative operating time to design life and the replacement cycle of key components; and the environmental trend characteristics are the slope of temperature and humidity changes in the past 7 days and environmental prediction data for the next 24 hours.

[0040] As a preferred solution for an AIoT-based alarm management device, the device association graph nodes in the root cause alarm information generation unit include: device, power supply line, environmental area, network switch, and the edges represent power supply association, network association, and location association.

[0041] As a preferred solution for an AIoT-based alarm management device, the operation and maintenance efficiency report and alarm trend report generation unit allocates work orders and tracks processing progress through the AIoT visual collaborative operation and maintenance platform according to the set allocation principles. Based on the principle of combining skill matching degree and proximity, work orders are allocated by combining the skill tags of operation and maintenance personnel, the distance to the fault point, and the current number of work orders.

[0042] The present invention has the following advantages:

[0043] First, alarm accuracy is significantly improved: By integrating multi-source data on equipment operation, environment, network and user operation, and combining the "random forest and isolated forest" combined algorithm, the system accurately distinguishes between valid and invalid alarms, reducing the false alarm rate from 25% of the existing technology to below 5% and the missed alarm rate to below 3%, greatly reducing the amount of invalid work orders processed by maintenance personnel.

[0044] Second, proactive prediction and accelerated fault response: Relying on bidirectional LSTM neural networks, combined with equipment historical faults, aging parameters and environmental trend data, potential faults can be predicted 1-4 hours in advance and alarms can be triggered in stages. The average time from alarm generation to fault resolution is reduced from 1.5 hours to within 15 minutes, significantly reducing production interruption losses.

[0045] Third, the operation and maintenance collaboration and management capabilities have been fully upgraded: Through the AIoT visual collaborative operation and maintenance platform, work orders are intelligently allocated based on the principle of "skill matching degree and nearest distance". Combined with progress tracking, historical query and statistical analysis functions, it covers the entire life cycle of alarms from "prediction-triggering-dispatch-processing-statistics", meeting the needs of enterprises for refined operation and maintenance management. At the same time, the work order processing results feed back into algorithm optimization, continuously improving system performance. Attached Figure Description

[0046] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0047] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0048] Figure 1 This is a flowchart illustrating an AIoT-based alarm management method provided in Embodiment 1 of the present invention.

[0049] Figure 2 This is a schematic diagram of the architecture of an AIoT-based alarm management device provided in Embodiment 2 of the present invention;

[0050] Figure 3 This is a schematic diagram of the specific composition architecture of an AIoT-based alarm management device provided in Embodiment 2 of the present invention. Detailed Implementation

[0051] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1

[0053] See Figure 1 Embodiment 1 of the present invention provides an alarm management method based on AIoT, comprising the following steps:

[0054] S1. Collect multi-dimensional basic data through AIoT terminals; obtain device history records and user operation trajectories by connecting to the device management system and user operation platform; preprocess the multi-dimensional basic data, device history records and user operation trajectories to obtain preprocessed data and upload it to the cloud platform; construct a multi-dimensional dataset based on the preprocessed data.

[0055] S2. By combining random forest and isolated forest algorithms, the multi-dimensional features in the multi-dimensional dataset are classified and calculated to identify valid alarm data and invalid alarm data.

[0056] S3. Based on the multi-dimensional dataset, combined with the valid alarm data and the invalid alarm data, the historical fault characteristics, equipment aging characteristics and environmental trend characteristics are analyzed by a bidirectional LSTM neural network, and the fault risk value for a specified period in the future is output. Based on the fault risk value, the corresponding level of pre-alarm is triggered.

[0057] S4. Based on the multi-dimensional dataset, combined with the valid alarm data and pre-alarm data, a device association graph is constructed through a graph neural network; when several devices generate valid alarms or pre-alarms at the same time, the abnormal state of the associated nodes is analyzed through the device association graph, the root cause of the fault is located and duplicate alarms are merged to generate root cause alarm information.

[0058] S5. Based on the root cause alarm information and pre-alarm level, work orders are allocated and processing progress is tracked through the AIoT visual collaborative operation and maintenance platform according to the set allocation principle; through the statistical analysis module, operation and maintenance efficiency reports and alarm trend reports are generated based on alarm data and work order processing data to realize alarm full lifecycle management.

[0059] In this embodiment, in step S1, multi-dimensional basic data is collected through an AIoT terminal; device history records and user operation trajectories are obtained by connecting to the device management system and the user operation platform; the multi-dimensional basic data, the device history records, and the user operation trajectories are preprocessed to obtain preprocessed data and uploaded to the cloud platform; and a multi-dimensional dataset is constructed based on the preprocessed data.

[0060] Specifically, the AIoT terminal includes three types of data acquisition components: device sensors are used to collect device operating parameters such as motor vibration, temperature, and voltage / current, with a collection frequency of once per second to ensure real-time capture of device operating status. Environmental sensors are used to monitor environmental data such as temperature, humidity, dust concentration, and air pressure, with temperature, humidity, and dust concentration collected once per minute, and air pressure collected once every 10 minutes, balancing data timeliness and acquisition cost. The network monitoring module acquires network parameters such as device network bandwidth, latency, and packet loss rate at a frequency of once per minute to ensure accurate network status monitoring. It interfaces with the device management system and user operation platform to obtain nearly 90 days of device maintenance records, fault logs, and nearly 24 hours of user operation data, including abnormal device start / stop and parameter modification records, excluding non-fault alarms caused by human error. Data preprocessing is performed through edge computing nodes, specifically including removing outliers exceeding normal parameter ranges and deleting duplicate data collected repeatedly within the same acquisition period. The preprocessed data is uploaded to the cloud platform and ultimately integrated into a multi-dimensional dataset containing real-time operational data, historical fault data, environmental data, network data, and user operation data, providing a comprehensive and clean data foundation for subsequent intelligent analysis.

[0061] In this embodiment, in step S2, the multi-dimensional features in the multi-dimensional dataset are classified and calculated using a combination algorithm of random forest and isolated forest to identify valid alarm data and invalid alarm data.

[0062] Specifically, four key features are first extracted from the multi-dimensional dataset: device features, environmental features, network features, and user features.

[0063] The equipment characteristics include: real-time vibration frequency, temperature, voltage, cumulative equipment runtime, and most recent maintenance time; environmental characteristics include temperature and humidity fluctuations and peak dust concentration over the past hour; network characteristics include average latency and packet loss rate over the past 30 minutes; and user characteristics include the number of abnormal operations over the past 24 hours.

[0064] These features are directly related to the determination of alarm authenticity. A combined algorithm of "random forest and isolation forest" is employed. Random forest is responsible for importance screening and preliminary classification of multi-dimensional features, identifying those that significantly impact alarm judgment and reducing interference from irrelevant features. Isolation forest then performs anomaly detection on the screened features, quickly identifying abnormal data deviating from the normal data distribution and corresponding potential alarms. The algorithm needs to be pre-trained using historical labeled data to achieve accurate classification of real-time data, ensuring a false alarm rate of ≤5% and a false negative rate of ≤3% for valid and invalid alarms, significantly reducing the effort required by operations personnel to handle invalid work orders.

[0065] In this embodiment, in step S3, based on the multi-dimensional dataset, combined with the valid alarm data and the invalid alarm data, a bidirectional LSTM neural network is used to analyze historical fault characteristics, equipment aging characteristics and environmental trend characteristics, output a fault risk value for a specified period in the future, and trigger a corresponding level of pre-alarm based on the fault risk value.

[0066] Specifically, the features required for the analysis are all derived from multi-dimensional datasets and previous alert data:

[0067] Historical fault characteristics include the fault occurrence time of similar equipment in the past 90 days in the multi-dimensional dataset and the trend of equipment parameter changes before the fault; equipment aging characteristics include the ratio of cumulative operating time to design life and the replacement cycle of key components; environmental trend characteristics include the slope of temperature and humidity changes in the past 7 days in the multi-dimensional dataset and environmental forecast data for the next 24 hours obtained from the meteorological platform.

[0068] Bidirectional LSTM neural networks have the advantage of capturing the temporal dependencies of data, and can simultaneously analyze the evolution of faults from historical data in both forward and reverse directions, ultimately outputting a fault risk value for the next 24 hours in the range of 0-100. The warning level is divided according to the risk value: a red warning is triggered when the risk value is ≥80 (high risk, requiring emergency handling), and a yellow warning is triggered when the risk value is ≥60 and <80 (medium risk, requiring close attention). The model prediction error rate is ≤10%, ensuring the reliability of the warning and allowing maintenance personnel 1-4 hours of advance handling time.

[0069] In this embodiment, in step S4, based on the multi-dimensional dataset, combined with the valid alarm data and the pre-alarm data, a device association graph is constructed through a graph neural network; when several devices generate valid alarms or pre-alarms at the same time, the abnormal state of the associated nodes is analyzed through the device association graph, the root cause of the fault is located and duplicate alarms are merged to generate root cause alarm information.

[0070] Specifically, a device association graph is constructed based on a graph neural network. The nodes of the graph include devices, power lines, environmental areas, and network switches. The edges between nodes represent the actual relationships. When multiple devices generate valid alarms or pre-alarms simultaneously, the system analyzes the abnormal state of the associated nodes through the device association graph to quickly locate the core abnormal node that caused the alarms from multiple devices. Subsequently, the alarms caused by the same core abnormal node from multiple devices are integrated into a single root cause alarm, avoiding repeated troubleshooting of multiple devices by maintenance personnel and significantly improving fault location efficiency.

[0071] In this embodiment, in step S5, based on the root cause alarm information and the pre-alarm level, work orders are allocated and processing progress is tracked through the AIoT visual collaborative operation and maintenance platform according to the set allocation principle; through the statistical analysis module, operation and maintenance efficiency reports and alarm trend reports are generated based on alarm data and work order processing data to realize alarm full lifecycle management.

[0072] Specifically, the AIoT visualized collaborative operation and maintenance platform supports both PC and mobile terminals, ensuring that operation and maintenance personnel can receive and process work orders anytime, anywhere. Work order allocation is based on the principles of "skill matching and proximity": skill matching is determined by matching the operation and maintenance personnel's skill tags with the fault type; proximity is determined by obtaining the real-time location of the operation and maintenance personnel's mobile device via GPS and allocating the optimal work order based on the personnel's current work order volume, while simultaneously pushing fault location, pre-diagnosis results, and handling suggestions; the work order processing progress is tracked in real time through the platform, with work order statuses including pending dispatch, received, arrived on-site, processing, and completed. Personnel can upload photos and records of on-site handling; the statistical analysis module generates reports based on full-process data: the operation and maintenance efficiency report includes data such as the number of work orders per person and the time to resolve faults; the alarm trend report is statistically analyzed by device type, region, and time dimension, and also supports querying historical alarm records and corresponding work order processing results by device number and time range; in addition, the work order processing results need to be fed back to the combined algorithm in step S2 and the bidirectional LSTM neural network in step S3 to optimize algorithm parameters, continuously improve system performance, and ultimately realize the full lifecycle management of alarms from "prediction-triggering-dispatch-processing-statistics".

[0073] In one possible embodiment, an example of a multi-motor operation and maintenance scenario in a heavy machinery production workshop is provided as follows:

[0074] The workshop is equipped with 10 production motors (numbered 1-10), and is equipped with the AIoT terminal, cloud platform, and collaborative operation and maintenance system described in this invention. The specific implementation steps are as follows:

[0075] T1. Data Acquisition and Preprocessing:

[0076] Multi-dimensional data collection: Each motor in the workshop is equipped with an AIoT terminal, which includes equipment sensors, environmental sensors, and a network monitoring module.

[0077] The equipment's sensors collect the vibration frequency (50±5Hz is within the normal range), stator temperature (≤80℃ is normal), and operating voltage / current (380V±10% voltage and 150±20A current are normal) of motors 1-10 in real time. The collection frequency is 1 time / second to ensure real-time capture of the motor's operating status.

[0078] Environmental sensors are deployed in the four corners of the workshop (east, west, south, and north) and the central area to monitor the workshop's temperature and humidity (temperature 18-28℃, humidity 40%-60% is normal), dust concentration (≤10mg / m³ is normal), and air pressure (standard atmospheric pressure ±5kPa is normal). The temperature, humidity, and dust concentration are collected once every 5 minutes, and the air pressure is collected once every 10 minutes, balancing data timeliness with terminal energy consumption.

[0079] The network monitoring module is integrated into the IoT gateway of each motor to obtain the bandwidth (≥10Mbps is normal), latency (≤50ms is normal), and packet loss rate (≤1% is normal) of the communication between the motor and the cloud platform. The sampling frequency is once per minute to ensure the accuracy of network communication status monitoring.

[0080] Historical data and operation trajectory acquisition: By connecting to the workshop equipment management system through the API interface, the maintenance records and fault logs of motors 1-10 for the past 90 days are obtained; at the same time, by connecting to the user operation platform, the equipment operation trajectory of workshop operation and maintenance personnel for the past 24 hours is obtained, including motor start and stop records and parameter modification records, to rule out the possibility of non-fault alarms caused by human operation.

[0081] Data preprocessing and dataset construction: Edge computing nodes perform preprocessing on the collected multi-dimensional data: data exceeding the normal parameter range is deleted, such as a voltage reading of 500V for motor 1 at a certain moment, which far exceeds the normal range of 380V±10%, and is determined to be instantaneous sensor fault data and discarded; identical data collected multiple times within the same collection cycle is deleted, such as temperature and humidity data collected continuously by environmental sensors at 10:00 and 10:00:02 being completely identical, the data from 10:00 is retained and duplicates are deleted; the preprocessed data is uploaded to the cloud platform via the 5G network, and finally integrated into a multi-dimensional dataset containing real-time motor operation data, environmental data, network data, historical maintenance fault data, and user operation data, providing clean and comprehensive data support for subsequent intelligent analysis.

[0082] T2. Alarm Identification and Filtering:

[0083] The cloud platform uses a combination of random forest and isolated forest algorithms to classify four key features in a multi-dimensional dataset, identifying valid and invalid alarms.

[0084] Extracting features related to alarm judgment from a multi-dimensional dataset:

[0085] Equipment characteristics: Real-time vibration frequencies of motors 1-3 are 52Hz, 53Hz, and 51Hz respectively; stator temperatures are 75℃, 76℃, and 74℃ respectively; operating voltage is 0V for all motors; cumulative running time is 6500 hours for motor 1, 6300 hours for motor 2, and 6400 hours for motor 3; last maintenance was performed 20 days ago for all motors.

[0086] Environmental characteristics: Temperature and humidity fluctuations in the workshop over the past hour: temperature fluctuation 1℃, humidity fluctuation 3%; peak dust concentration 8mg / m³, all within the normal range;

[0087] Network characteristics: The average latency of motors 1-3 over the past 30 minutes was 60ms; the packet loss rate was 100% for all three motors, exceeding the normal range.

[0088] User characteristics: There were no abnormal operation records for motors 1-3 in the past 24 hours, and the number of abnormal operations was 0.

[0089] Algorithm classification calculation:

[0090] The random forest algorithm first filters the importance of the above four types of features, and determines that "operating voltage 0V" and "packet loss rate 100%" are the core features that have the greatest impact on alarm judgment, and removes irrelevant features such as "ambient temperature and humidity fluctuations" and "number of user operations".

[0091] The Isolation Forest algorithm performs anomaly detection based on the core features after screening. It found that the data combination of "0V voltage and 100% packet loss rate" of motors 1-3 deviated significantly from the normal data distribution. Combined with historical labeled data, the "disconnection alarm" reported by motors 1-3 was finally identified as valid alarm data, and no invalid alarms were generated, ensuring that the false alarm rate of the alarm screening results is ≤5%.

[0092] T3. Fault Alarm Analysis:

[0093] Based on a multi-dimensional dataset and combined with the valid alarm data identified in step T2, the cloud platform analyzes fault-related features using a bidirectional LSTM neural network, outputs a fault risk value, and triggers a pre-alarm.

[0094] Three key features are input into the bidirectional LSTM neural network:

[0095] Historical fault characteristics: Multi-dimensional data collection of fault records of motors of the same type (models 4-6, which are the same as motors 1-3) over the past 90 days shows that this type of motor experienced a power outage fault again within 72 hours after "abnormal power supply line voltage", and the number of power supply line voltage fluctuations before the fault increased by 25%-35% compared with the normal period.

[0096] Equipment aging characteristics: The cumulative operating time of the power supply line L1 associated with motors 1-3 is 5000 hours, the design life is 6000 hours, the aging rate is 5000 / 6000=83%, and the replacement cycle of the connector components of L1 is 4500 hours, which has now exceeded the service life by 500 hours.

[0097] Environmental trend characteristics: The slope of temperature and humidity changes in the workshop over the past 7 days was collected from the multi-dimensional dataset and connected with the environmental forecast data for the next 24 hours obtained from the local meteorological platform. At the same time, the voltage data of L1 over the past 3 months was extracted, and it was found that the number of voltage fluctuations increased by 30% compared with the previous 3 months.

[0098] Risk value output and early warning triggering: By analyzing the temporal dependencies of the above features, the bidirectional LSTM neural network outputs a fault risk value of 85 for power supply line L1 in the next 24 hours. According to the risk value classification rules, a risk value ≥80 triggers a red early warning. At the same time, combined with the historical fault recurrence cycle, it further predicts that L1 may lose power again in the next 72 hours, triggering a yellow early warning. The model prediction error rate is ≤10%, allowing sufficient time for maintenance personnel to handle the situation.

[0099] T4. Alarm Correlation Analysis and Root Cause Location:

[0100] Based on a multi-dimensional dataset, the cloud platform combines the valid alarm data from step T2 with the pre-alarm data from step T3, and constructs a device association map using a graph neural network (GNN) and performs association analysis.

[0101] Equipment Association Graph Construction: The graph nodes include motors 1-10, power supply lines L1-L3, where motors 1-3 are associated with L1, motors 4-6 are associated with L2, motors 7-10 are associated with L3, workshop environment areas (East / West / South / North / Central), and network switches S1-S2. Motors 1-5 are connected to S1, and motors 6-10 are connected to S2. The edges between nodes represent the actual association relationships: motors 1-3 are associated with L1 through "power supply", motors 1-3 are associated with S1 through "network", and motors 1-3 are associated with the central area of ​​the workshop through "location".

[0102] Correlation Analysis and Root Cause Location: When motors 1-3 simultaneously report "disconnection alarms," ​​the system analyzes the abnormal status of related nodes using an equipment correlation graph.

[0103] The voltage monitoring value of power supply line L1 is 0V, which is obviously abnormal;

[0104] All operating parameters of network switch S1 are normal, ruling out network device failure;

[0105] The temperature and humidity in the central area of ​​the workshop were normal, ruling out environmental factors. Based on the above analysis, the root cause of the fault was determined to be "power outage of L1 power supply line". The "disconnection alarms" of motors 1-3 were all derivative alarms caused by the power outage of L1. The system then merged the three "disconnection alarms" into one "root cause alarm: power outage of L1 power supply line, affecting the operation of motors 1-3, risk level red", avoiding the need for maintenance personnel to check the three motors separately and improving the efficiency of fault location.

[0106] T5. Collaborative Operation and Maintenance and Full Lifecycle Management:

[0107] The AIoT visualized collaborative operation and maintenance platform, based on the root cause alarm information in step T4 and the pre-alarm level in step T3, performs work order scheduling, progress tracking, and statistical analysis to achieve full lifecycle management of alarms.

[0108] Intelligent work order allocation: The platform connects to the mobile GPS devices of 5 maintenance personnel in the workshop to obtain their real-time location and skill tags.

[0109] Engineer Wang: Skill tags "Power Line Repair" and "Motor Repair", currently in the central area of ​​the workshop; Engineer Li: Skill tag "Network Debugging", currently in the east gate area of ​​the workshop; the other 3 maintenance personnel all have unprocessed work orders;

[0110] Based on the principle of "skill matching and proximity", Wang Gong's "power line maintenance" tag is a perfect match for the "L1 power outage" fault type, and his current location is only 30 meters away from the L1 fault point, making him the best candidate.

[0111] The work order is automatically generated and pushed to Wang Gong's mobile device. The work order content includes: "Root cause alarm: Power supply line L1 is out of power, risk level red; Warning: L1 may be out of power again in the next 72 hours, risk level yellow; Suggested handling: Check the break point of L1 line from the power distribution room to motor section 1-3; Replace the overdue joint components of L1 line; Test the line voltage stability after repair."

[0112] Work order progress tracking:

[0113] At 10:05, Engineer Wang received the work order, and the platform updated the work order status to "received"; at 10:08, he arrived at the L1 fault point and clicked "arrived at the site" on his mobile device, and the status was updated synchronously.

[0114] According to the work order suggestions, the following troubleshooting steps were taken: At 10:10, it was found that the connector of the L1 line near motor 2 was aged and broken. At 10:15, the connector part of model RVV3×10 was replaced. At 10:20, the line voltage was tested and restored to 220V. Motors 1-3 were reconnected to the network. At the same time, the entire L1 line was tested and another connector near motor 3 was found to show signs of aging. At 10:30, the connector was replaced and the risk of the early warning was eliminated.

[0115] At 10:35, Engineer Wang uploaded photos and records of the on-site handling to the mobile device and clicked "Complete Work Order". The platform updated the status to "Completed". The entire fault handling took 30 minutes.

[0116] Statistical analysis and algorithm optimization:

[0117] Based on the alarm data and work order processing data, the statistical analysis module automatically updates the maintenance efficiency report: Engineer Wang processed a total of 12 work orders this month, with an average fault resolution time of 28 minutes, which is lower than the workshop average of 35 minutes; at the same time, the alarm trend report is updated: there were a total of 2 power line alarms this month, both concentrated on the L1 line, and the central area of ​​the workshop accounted for 30% of the alarms by region.

[0118] The platform supports querying alarm records and work order processing results by "Equipment Number = Motors 1-3" and "Time Range = Today 10:00-10:40", facilitating subsequent traceability;

[0119] The system feeds back the processing results to the "random forest and isolated forest" combined algorithm in step T2 and the bidirectional LSTM neural network in step T3 to update the feature weights of "power line aging fault" and continuously optimize the alarm identification and fault prediction accuracy of the algorithm.

[0120] As can be seen from the above embodiments, the present invention can achieve accurate identification of alarms, early prediction of faults, rapid location of root causes, and efficient collaborative operation and maintenance in the operation and maintenance of multiple devices in industrial workshops, significantly reducing production interruption losses and meeting the needs of enterprises for refined operation and maintenance.

[0121] The application scenarios of this invention are as follows:

[0122] When applied to multi-equipment cluster operation and maintenance scenarios in industrial workshops, this invention collects operating data of equipment such as motors and machine tools, workshop environmental data, and network data through AIoT terminals. Combining historical equipment fault records with user operation trajectories, it accurately identifies valid alarms through a combination algorithm, predicts potential faults using an LSTM model, locates the root cause of faults using a GNN graph, and then intelligently dispatches work orders through a collaborative operation and maintenance platform. This can quickly resolve the root cause problem of multiple devices alarming simultaneously, significantly shorten fault handling time, and reduce production interruption losses.

[0123] When applied to smart park infrastructure operation and maintenance scenarios, this invention can collect operating parameters of infrastructure such as power distribution equipment, HVAC systems, and security cameras within the park through AIoT terminals, connect to the park management system to obtain equipment maintenance records, and, through data preprocessing and multi-dimensional analysis, predict risks such as aging power distribution lines and HVAC equipment failures in advance. When multiple areas of equipment alarm simultaneously, the root cause problem can be located through correlation maps, and work orders can be automatically dispatched based on the location and skills of operation and maintenance personnel, thereby realizing full lifecycle management of park infrastructure alarms and improving park operation and maintenance efficiency.

[0124] When applied to the operation and maintenance of equipment on smart factory production lines, this invention can collect real-time operating data of equipment such as production line conveyors, robotic arms, and sensors, obtain fault logs and maintenance records of the equipment for nearly 90 days, eliminate invalid alarms caused by transient signal interference through "random forest and isolated forest" algorithms, predict potential faults such as wear of robotic arm bearings and conveyor belt jamming using a bidirectional LSTM model, merge duplicate information to generate root cause alarms when multiple devices alarm, and then push work orders to nearby operation and maintenance personnel with corresponding skills through a collaborative platform to ensure the continuous and stable operation of the production line.

[0125] When applied to data center equipment operation and maintenance scenarios, this invention can collect operating data of servers, switches, and air conditioning systems through AIoT terminals, connect to the data center management platform to obtain historical operation and maintenance records of the equipment, accurately identify effective alarms such as server disconnection and switch failure, predict risks such as insufficient cooling of the air conditioning system and aging of server hardware, locate the root cause of the fault when multiple devices alarm, allocate work orders according to the principle of "skill matching and nearest distance", shorten the data center fault recovery time, and ensure the security of data storage and transmission.

[0126] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0127] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0128] Example 2

[0129] See Figure 2 and Figure 3 Embodiment 2 of the present invention also provides an alarm management device based on AIoT, comprising:

[0130] The multi-dimensional dataset construction unit 001 is used to collect multi-dimensional basic data through AIoT terminals; obtain device history records and user operation trajectories by connecting to the device management system and the user operation platform; preprocess the multi-dimensional basic data, the device history records and the user operation trajectories to obtain preprocessed data and upload it to the cloud platform; and construct a multi-dimensional dataset based on the preprocessed data.

[0131] The valid alarm data and invalid alarm data identification unit 002 is used to classify and calculate the multi-dimensional features in the multi-dimensional dataset using a combination algorithm of random forest and isolated forest, and to identify valid alarm data and invalid alarm data.

[0132] The early warning unit 003 is used to analyze historical fault characteristics, equipment aging characteristics and environmental trend characteristics based on the multi-dimensional dataset, combined with the valid alarm data and the invalid alarm data, through a bidirectional LSTM neural network, output a fault risk value for a specified period in the future, and trigger a corresponding level of early warning based on the fault risk value.

[0133] The root cause alarm information generation unit 004 is used to construct a device association map through a graph neural network based on the multi-dimensional dataset, combined with the valid alarm data and the pre-alarm data; when several devices generate valid alarms or pre-alarms at the same time, the abnormal state of the associated nodes is analyzed through the device association map, the root cause of the fault is located and duplicate alarms are merged to generate root cause alarm information.

[0134] The operation and maintenance efficiency report and alarm trend report generation unit 005 is used to allocate work orders and track processing progress through the AIoT visual collaborative operation and maintenance platform according to the set allocation principles based on the root cause alarm information and pre-alarm level; and to generate operation and maintenance efficiency reports and alarm trend reports based on alarm data and work order processing data through the statistical analysis module, so as to realize alarm full life cycle management.

[0135] In this embodiment, the multi-dimensional dataset construction unit 001 includes the following multi-dimensional basic data: motor vibration, temperature, voltage / current collected by device sensors; temperature, humidity, dust concentration, and air pressure monitored by environmental sensors; and bandwidth, latency, and packet loss rate obtained by the network monitoring module.

[0136] The equipment history record includes equipment maintenance records and fault logs for the past 90 days.

[0137] The user operation trajectory is a record of abnormal device start-ups and shutdowns and parameter modifications over the past 24 hours.

[0138] During the preprocessing of the multi-dimensional basic data, the device historical records, and the user operation trajectory, outlier removal and duplicate data deletion are performed through edge computing nodes to obtain the preprocessed data.

[0139] In this embodiment, in the valid alarm data and invalid alarm data identification unit 002, during the process of classifying and calculating the multi-dimensional features in the multi-dimensional dataset, the multi-dimensional features include: device features, environmental features, network features and user features;

[0140] The equipment features include real-time vibration frequency, temperature, voltage, cumulative operating time, and most recent maintenance time;

[0141] The environmental characteristics are the temperature and humidity fluctuation values ​​and the peak dust concentration over the past hour.

[0142] The network characteristics are the average latency and packet loss rate over the past 30 minutes;

[0143] The user characteristics are the number of abnormal operations in the past 24 hours.

[0144] In this embodiment, in the early warning unit 003, during the analysis of the historical fault characteristics, the equipment aging characteristics, and the environmental trend characteristics through the bidirectional LSTM neural network, the historical fault characteristics are the occurrence time of similar equipment faults and the trend of parameter changes before the fault in the past 90 days; the equipment aging characteristics are the ratio of cumulative running time to design life and the replacement cycle of key components; and the environmental trend characteristics are the slope of temperature and humidity changes in the past 7 days and environmental prediction data for the next 24 hours.

[0145] In this embodiment, the device association graph nodes in the root cause alarm information generation unit 004 include: device, power supply line, environmental area, network switch, and the edges represent power supply association, network association and location association.

[0146] In this embodiment, the operation and maintenance efficiency report and alarm trend report generation unit 005 allocates work orders and tracks processing progress through the AIoT visual collaborative operation and maintenance platform according to the set allocation principle. Based on the principle of combining skill matching degree and proximity, work orders are allocated in combination with the skill tags of operation and maintenance personnel, the distance of the fault point and the current number of work orders.

[0147] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0148] Example 3

[0149] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code of an AIoT-based alarm management method. The program code includes instructions for executing the AIoT-based alarm management method of Embodiment 1 or any possible implementation thereof.

[0150] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0151] Example 4

[0152] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0153] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute an AIoT-based alarm management method according to Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0154] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0155] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0156] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0157] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. An AIoT-based alarm management method, characterized in that, The application relates to an AIoT terminal-based multi-dimensional data collection method and system. The method comprises the following steps: Collecting multi-dimensional basic data through an AIoT terminal; Obtaining device historical records and user operation trajectories through a device management system and a user operation platform; Preprocessing the multi-dimensional basic data, the device historical records and the user operation trajectories to obtain preprocessed data and uploading the preprocessed data to a cloud platform; Based on the preprocessed data, a multi-dimensional data set is constructed; Through a combination algorithm of random forests and isolated forests, multi-dimensional features in the multi-dimensional data set are classified and calculated to identify effective alarm data and invalid alarm data; Based on the multi-dimensional data set, the effective alarm data and the invalid alarm data are combined, historical fault features, device aging features and environmental trend features are analyzed through a bidirectional LSTM neural network, future specified period fault risk values are output, and corresponding level pre-alarm is triggered according to the fault risk values; Based on the multi-dimensional data set, the effective alarm data and the pre-alarm data are combined, a device correlation graph is constructed through a graph neural network; when a plurality of devices simultaneously generate effective alarms or pre-alarm, the device correlation graph is used to analyze the abnormal state of the correlation nodes, the fault root cause is located and repeated alarms are merged to generate root alarm information; Based on the root alarm information and the pre-alarm level, work orders are distributed and processing progress is tracked through an AIoT visual collaborative operation and maintenance platform according to a set distribution principle; 2.The AIoT-based alarm management method of claim 1, wherein, Through a statistical analysis module, operation and maintenance efficiency reports and alarm trend reports are generated based on alarm data and work order processing data, and alarm life cycle management is realized. The multi-dimensional basic data comprises motor vibration, temperature, voltage / current collected by a device sensor, temperature and humidity, dust concentration, air pressure monitored by an environmental sensor, bandwidth, delay and packet loss rate obtained by a network monitoring module; The device historical records are device maintenance records and fault logs in the past 90 days; The user operation trajectory is a record of abnormal start-stop and parameter modification of a device in the past 24 hours; 3.The AIoT-based alarm management method of claim 2, wherein, In the preprocessing process of the multi-dimensional basic data, the device historical records and the user operation trajectory, outlier elimination and repeated data deletion are performed through an edge computing node to obtain the preprocessed data. In the classification and calculation process of the multi-dimensional features in the multi-dimensional data set, the multi-dimensional features comprise device features, environmental features, network features and user features; The device features are real-time vibration frequency, temperature, voltage, cumulative running time and recent maintenance time; The environmental features are temperature and humidity fluctuation values in the past 1 hour and dust concentration peak values; The network features are average delay and packet loss rate in the past 30 minutes; The user features are abnormal operation times in the past 24 hours. 4.The AIoT-based alarm management method of claim 3, wherein, In the process of analyzing the historical failure features, the equipment aging features and the environmental trend features by the bidirectional LSTM neural network, the historical failure features are the same type of equipment failure time and parameter change trend before failure in the past 90 days; the equipment aging features are the cumulative running time / design life ratio and key component replacement cycle; and the environmental trend features are the temperature and humidity change slope in the past 7 days and the future 24-hour environmental prediction data. 5.The AIoT-based alarm management method of claim 4, wherein, The device association graph node comprises a device, a power supply line, an environmental area and a network switch, and the edges represent power supply association, network association and location association. 6.The AIoT-based alarm management method of claim 5, wherein, In the process of distributing work orders and tracking processing progress by the AIoT visual collaborative operation and maintenance platform according to the set distribution principle, the work orders are distributed based on the principle of combining skill matching degree and nearest distance, in combination with the skill labels of operation and maintenance personnel, the distance of fault points and the current work order quantity.

7. An AIoT-based alarm management apparatus, adopting an AIoT-based alarm management method according to any one of claims 1-6, characterized in that, It comprises: A multi-dimensional data set construction unit for collecting multi-dimensional basic data through an AIoT terminal; By interfacing with a device management system and a user operation platform, device historical records and user operation trajectories are obtained; the multi-dimensional basic data, the device historical records and the user operation trajectories are preprocessed to obtain preprocessed data and upload the preprocessed data to a cloud platform; Based on the preprocessed data, a multi-dimensional data set is constructed; An effective alarm data and invalid alarm data identification unit for classifying and calculating multi-dimensional features in the multi-dimensional data set by a combination algorithm of random forest and isolated forest to identify effective alarm data and invalid alarm data; A pre-alarm unit for analyzing historical failure features, equipment aging features and environmental trend features by a bidirectional LSTM neural network based on the multi-dimensional data set in combination with the effective alarm data and the invalid alarm data, outputting a future specified period failure risk value, and triggering a corresponding level pre-alarm according to the failure risk value; A root cause alarm information generation unit for constructing a device association graph by a graph neural network based on the multi-dimensional data set in combination with the effective alarm data and pre-alarm data; when a plurality of devices simultaneously generate effective alarms or pre-alarms, the device association graph is used to analyze the abnormal state of the associated nodes, locate the fault root cause and merge repeated alarms to generate root cause alarm information; An operation efficiency report and alarm trend report generation unit for distributing work orders and tracking processing progress by an AIoT visual collaborative operation and maintenance platform according to a set distribution principle based on the root cause alarm information and pre-alarm level; Through a statistical analysis module, an operation efficiency report and an alarm trend report are generated based on alarm data and work order processing data to realize alarm life cycle management. 8.The AIoT-based alarm management apparatus of claim 7, wherein, In the multi-dimensional data set construction unit, the multi-dimensional basic data comprises motor vibration, temperature, voltage / current collected by a device sensor, temperature and humidity, dust concentration, air pressure monitored by an environmental sensor, and bandwidth, delay and packet loss rate obtained by a network monitoring module; The device historical records are device maintenance records and fault logs in the past 90 days; The user operation trajectory is a record of abnormal start and stop of the device and parameter modification in the past 24 hours. In the process of preprocessing the multi-dimensional basic data, the device history record and the user operation trajectory, outlier elimination and duplicate data deletion are performed by an edge computing node to obtain the preprocessed data. 9.The AIoT-based alarm management apparatus of claim 8, wherein, In the process of classifying and calculating the multi-dimensional features in the multi-dimensional data set in the effective and ineffective alarm data identification unit, the multi-dimensional features include device features, environment features, network features and user features. The device features are real-time vibration frequency, temperature, voltage, cumulative running time and recent maintenance time. The environment features are near 1-hour temperature and humidity fluctuation value and dust concentration peak value. The network features are near 30-minute average delay and packet loss rate. The user features are near 24-hour abnormal operation times. 10.The AIoT-based alarm management apparatus of claim 9, wherein, In the process of analyzing the historical failure features, the device aging features and the environment trend features by the bidirectional LSTM neural network in the pre-alarm unit, the historical failure features are the occurrence time of the same type of device failure and the parameter change trend before the failure in the past 90 days; the device aging features are the cumulative running time / design life ratio and the key component replacement cycle; and the environment trend features are the temperature and humidity change slope in the past 7 days and the future 24-hour environment prediction data. In the process of generating the root cause alarm information in the root cause alarm information generation unit, the device association graph nodes include devices, power supply lines, environment areas, network switches, and edges representing power supply association, network association and location association. In the process of distributing work orders and tracking processing progress by the AIoT visual collaborative operation and maintenance platform according to the set distribution principle in the operation and maintenance efficiency report and alarm trend report generation unit, work orders are distributed based on the principle of combining skill matching degree and distance proximity, combined with operation and maintenance personnel skill labels, fault point distance and current work order quantity.

Images