Personnel positioning node fault rapid diagnosis method and system

Abstract

The invention discloses a rapid fault diagnosis method and system for a personnel positioning node, relates to the technical field of fault diagnosis, and aims to solve the problems that a current positioning node fault diagnosis method has single index limitation and lacks time sequence prediction. The method comprises the following steps: S1, collecting multi-source data and preprocessing the multi-source data to generate multi-dimensional features; s2, inputting the multi-dimensional features into a trained LSTM prediction model according to a feature adaptive selection mechanism, wherein output values are an RSSI prediction value and a TOF prediction value; s3, according to the output value, the fault type is judged through priority ranking and a time window verification rule set; s4, grading the faults, and executing alarm according to the constructed multi-stage alarm mechanism; the invention further comprises a corresponding system. According to the personnel positioning node fault rapid diagnosis method, rapid diagnosis of the personnel positioning node fault can be realized, and the diagnosis accuracy and the fault prediction capability can be improved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to a method and system for rapid fault diagnosis of personnel positioning nodes. Background Technology

[0002] In oil and gas operation scenarios, personnel positioning nodes are prone to failure due to complex environments (such as electromagnetic interference, vibration, and dust). Existing diagnostic methods face three core challenges: 1. Environmental complexity leads to diverse failure modes: High temperatures (not less than 60℃), high humidity (not less than 90%), flammable and explosive gases (such as methane), and strong electromagnetic interference (such as pumping unit frequency converters) can cause complex failures such as clock drift in positioning modules, short circuits in communication module antennas, and thermal runaway in power modules. Traditional single-index diagnostic methods have a false alarm rate exceeding 30%. 2. Conflict between real-time requirements and computing resources: Emergency scenarios such as well blowouts require a fault response time of less than 100ms, but existing deep learning algorithms (such as CNN) have a single-node inference time of more than 500ms, which cannot meet the needs of edge computing. 3. Conflict between fixed thresholds and dynamic environmental changes: Seasonal temperature differences in oil and gas fields (-20℃ to 60℃) can cause RSSI threshold drift of ±15dBm. Fixed thresholds result in a false alarm rate as high as 45% in summer.

[0003] For example, Chinese patent CN120141443A relates to a method for locating people based on multi-source data, but this patent still cannot solve the three core challenges mentioned above. Summary of the Invention

[0004] This invention addresses the limitations of current location node fault diagnosis methods, which rely on a single indicator and lack time-series prediction. It proposes a rapid fault diagnosis method and system for personnel location nodes, which can achieve rapid fault diagnosis of personnel location nodes and improve diagnostic accuracy and fault prediction capabilities.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for rapid diagnosis of personnel positioning node faults, comprising the following steps: S1, collect multi-source data and preprocess the multi-source data to generate multi-dimensional features; S2, input the multidimensional features into the trained LSTM prediction model according to the feature adaptive selection mechanism, and output the RSSI prediction value and the TOF prediction value. S3, based on the output value, uses priority sorting and time window verification rule set to determine the fault type; S4 classifies faults and executes alarms based on the constructed multi-level alarm mechanism.

[0006] The core of this invention lies in building an efficient and reliable rapid fault diagnosis system for oil and gas operation personnel positioning nodes by integrating a multi-dimensional feature fusion diagnostic architecture, a lightweight LSTM edge computing model, a dynamic rule adaptive mechanism, and a three-level fault classification alarm system. This system enables accurate diagnosis and fault prediction of personnel positioning nodes.

[0007] The present invention is further configured such that step S1 includes: acquiring several types of raw data respectively, performing analog-to-digital conversion through an ADC module, and transmitting the data directly to the MCU after the conversion is completed; then performing preprocessing on the data, including outlier filtering, normalization processing and timing alignment processing.

[0008] In this technical solution, the TOF raw data is collected in real time through the positioning module SPI interface, the RSSI sequence is obtained through the communication module MAC layer, the battery voltage is read through the power chip, and the node ambient temperature is collected through the thermistor battery.

[0009] The present invention is further configured such that the feature adaptive selection mechanism includes: A preset feature priority mapping table is used to divide features into core features, auxiliary features, and supplementary features based on the correlation between the fault type and features in the oil and gas scenario. When in basic diagnosis mode, only core features are enabled. When in precise diagnosis mode, auxiliary features are activated if the anomaly score of the core features is not less than 0.2. Supplementary features are added when in difficult diagnosis mode.

[0010] In this technical solution, the generated multidimensional features are divided, and the corresponding features or feature combinations are activated according to different situations.

[0011] The present invention is further configured such that step S2 includes: S21, an LSTM prediction model is constructed based on a two-layer recurrent neural network structure, with the input layer being a temporal feature matrix; S22, determine the specific features and feature dimensions input to the LSTM prediction model based on the feature adaptive selection mechanism; S23, the fully connected layer of the LSTM prediction model outputs the RSSI prediction value and TOF prediction value for the next time step.

[0012] In this technical solution, the input layer of the LSTM prediction model adopts a flexible dimension design. Through the feature dimension standardization module, the input features of different dimensions are uniformly mapped to a 100×N time series matrix (N is the selected feature dimension). It can adapt to any dimension of N=4~8 without modifying the core network structure.

[0013] The present invention is further configured such that step S23 includes calculating anomaly scores, specifically: calculating the normalized error between the measured RSSI value and the predicted RSSI value, and the normalized error between the measured TOF value and the predicted TOF value, denoted as […]. and .

[0014] In this technical solution, the received signal strength indication directly reflects the working status of the communication module, and the flight time directly reflects the working status of the positioning module. Together, they cover more than 80% of node failure scenarios.

[0015] The present invention is further configured to: determine the fault type using a priority sorting and time window verification rule set, including: The rules are categorized and sorted according to preset priority, and the triggering conditions of each rule are verified in descending order of priority. For rules that meet the triggering conditions, verify whether the duration of the abnormal feature reaches a preset threshold; When multiple rules of the same priority meet the triggering condition, record the timestamp of the first time each rule meets the triggering condition, and select the rule with the latest timestamp as the current diagnostic result.

[0016] In this technical solution, priority strategy and time window strategy are used to execute the set of rules.

[0017] The present invention is further configured such that: the rules include rule categories, triggering conditions, fault types and priorities, and all rules are stored to form a rule set.

[0018] In this technical solution, the rule set can be displayed in tabular form.

[0019] The present invention is further configured such that: the multi-level alarm mechanism includes a three-level alarm system, the three-level alarm system including: When the fault type is a potential anomaly, a Level 1 early warning is triggered; when the fault is a confirmed fault, a Level 2 alarm is triggered; when the fault involves a safety threat, a Level 3 emergency shutdown is initiated, cutting off the power supply to the relevant equipment.

[0020] In this technical solution, three levels of alarms are defined, which enables more accurate decision-making and execution.

[0021] The present invention is further configured such that: the core feature is a four-dimensional feature, the auxiliary feature is a two-dimensional feature, and the supplementary feature is a two-dimensional feature.

[0022] A rapid fault diagnosis system for personnel positioning nodes, applicable to the aforementioned rapid fault diagnosis method for personnel positioning nodes, includes: The data acquisition module collects data from multiple sources. The preprocessing module performs preprocessing on multi-source data; The predictive analysis module uses an LSTM prediction model for prediction. The decision execution module includes fault type identification and alarm execution.

[0023] In this technical solution, the acquisition module is connected to the preprocessing module, the preprocessing module is connected to the predictive analysis module, and the predictive analysis module is connected to the decision execution module. The acquisition module, preprocessing module, predictive analysis module, and decision execution module can work together to perform fault diagnosis and fault prediction of personnel positioning nodes.

[0024] The present invention can bring the following beneficial effects: This application relates to a rapid fault diagnosis method for personnel positioning nodes. By integrating a multi-dimensional feature fusion diagnostic architecture, a lightweight LSTM edge computing model, a dynamic rule adaptive mechanism, and a three-level fault classification alarm system, it can perform fault diagnosis and fault prediction for personnel positioning nodes, and can significantly improve the diagnostic accuracy and fault prediction capability. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for rapid diagnosis of personnel positioning node faults according to this application.

[0026] Figure 2 This is a detailed architecture diagram of an LSTM model for a rapid fault diagnosis method for personnel positioning nodes proposed in this application. Detailed Implementation

[0027] Example 1 This embodiment proposes a rapid diagnosis method for personnel positioning node faults, referencing... Figure 1 and Figure 2 It mainly includes the following steps.

[0028] Step S1: Collect multi-source data, then preprocess the collected multi-source data to obtain multi-dimensional features.

[0029] For step S1, the process is as follows: multiple types of raw data are collected, and then analog-to-digital conversion is performed by the ADC module. After the conversion is completed, the converted data is transmitted to the MCU. Then, the data is preprocessed. The preprocessing process includes, but is not limited to, outlier filtering, normalization and timing alignment.

[0030] In this embodiment, eight types of data are collected, namely Received Signal Strength Indicator (RSSI), Time of Flight (TOF), battery voltage, temperature, operating mode, communication quality, signal multipath index, and node load rate.

[0031] For example, the TOF raw data is acquired in real time through the positioning module's SPI interface, with a sampling frequency of 10Hz; the RSSI sequence is obtained through the communication module's MAC layer, with a sampling frequency of 10Hz; the battery voltage is read through the power chip, with a sampling frequency of 0.5Hz; and the node's ambient temperature is acquired through the thermistor, with a sampling frequency of 0.1Hz.

[0032] In this embodiment, a low-power ADC (12-bit resolution) is used to convert the analog signal and transmit it directly to the MCU (STM32H743) through the DMA channel to avoid CPU interrupt delay.

[0033] Data preprocessing includes the following steps.

[0034] Outlier filtering: The RSSI sequence is processed using the 3σ principle, which is: the threshold is equal to μ plus or minus 3σ, where μ is the mean and σ is the standard deviation. When the RSSI exceeds the threshold at a certain moment, the outlier is replaced by forward interpolation (xn=xn-1+(xn+1-xn-1) / 2).

[0035] Normalization: For data with different dimensions such as TOF (range 0~100m) and voltage (3.0~4.2V), minimum-maximum normalization is used.

[0036] Timing alignment: Data in each dimension is synchronized by timestamps. Zero-order hold interpolation is used for signals with different sampling frequencies (such as voltage 0.5Hz and RSSI 1Hz) to unify them to a sampling rate of 1Hz.

[0037] For personnel positioning nodes, they are hardware devices deployed in oil and gas operation areas (such as drilling platforms and along oil pipelines), integrating positioning modules (UWB / BeiDou), communication modules (LoRa / WiFi), and power modules (lithium batteries + energy harvesters). They achieve personnel location tracking through signal interaction, with a typical deployment density of one node every 50 meters.

[0038] Step S2: The multidimensional features generated in step S1 are input into the trained LSTM prediction model according to the feature adaptive selection mechanism, and the RSSI prediction value and TOF prediction value are output. RSSI is the Received Signal Strength Indicator, a physical quantity that quantifies the signal transmission strength between positioning nodes, measured in dBm. In oil and gas environments, it is significantly affected by reflections from metal pipes and electromagnetic interference, and its normal range is typically -40dBm to -80dBm. TOF is the Time of Flight, which calculates the distance between nodes by measuring the time it takes for electromagnetic waves to travel from transmission to reception. Positioning accuracy can reach the centimeter level, but it is affected by ambient temperature and humidity, resulting in transmission delay errors.

[0039] The feature adaptive selection mechanism is as follows: First, a feature priority mapping table is set in advance. Based on the correlation between the fault type of oil and gas scenario and the feature, different features are divided into core features, auxiliary features and supplementary features. In this embodiment, the core feature is a four-dimensional feature, and the auxiliary features and supplementary features are both two-dimensional features.

[0040] The specific dynamic selection of trigger conditions is as follows: in the basic diagnostic mode, only the core features need to be enabled; in the precise diagnostic mode, the auxiliary features are activated when the abnormal score of the core features is greater than or equal to 0.2; and in the difficult diagnostic mode, supplementary features are added.

[0041] In this embodiment, the core feature is a four-dimensional feature, which is an example of a preferred embodiment and is not a limitation: the input feature dimension of the LSTM model of the present invention can be flexibly extended to five-dimensional, six-dimensional or even eight-dimensional (received signal strength indication, flight time, battery voltage, temperature, working mode, communication quality, signal multipath index, node load rate), depending on the accuracy requirements of the fault diagnosis scenario and the upper limit of the computing resources of the edge nodes.

[0042] Specifically, for the feature adaptive selection mechanism, this embodiment pre-defines a feature priority mapping table: based on the correlation between the fault type in the oil and gas scenario and the features, core features (in this embodiment, these include received signal strength indication, flight time, voltage, and temperature); auxiliary features (communication quality, signal multipath index); and supplementary features (operating mode, node load rate) are divided into core features, auxiliary features, and supplementary features, with the priority decreasing sequentially.

[0043] When in basic diagnostic mode (e.g., when edge node computing power is limited): only 3D core features are enabled to ensure inference time is less than 50ms.

[0044] When in precise diagnostic mode (e.g., core work area nodes with sufficient computing power): when the anomaly score of the core feature is ≥0.2, auxiliary features are automatically activated, expanding the input dimension to six dimensions to improve the accuracy of composite fault identification. When in difficult diagnosis mode (e.g., the fault type is not clear for 3 consecutive time steps): further supplementary features are added to expand to eight-dimensional full feature input, and the root cause of the fault is locked through multi-dimensional cross-validation.

[0045] In this embodiment, a feature dimension adaptation mechanism is adopted. Specifically, the model input layer adopts a flexible dimension design. Through the feature dimension standardization module, the input features of different dimensions are uniformly mapped to a 100×N time series matrix (N is the selected feature dimension). It can adapt to any dimension of N=4~8 without modifying the core network structure.

[0046] Step S2 mainly includes the following process.

[0047] Step S21: Based on the two-layer recurrent neural network structure, an LSTM prediction model is established, with the input layer of the model being a temporal feature matrix.

[0048] For step S21, refer to Figure 2 The LSTM prediction model in this embodiment uses a two-layer recurrent neural network structure. The input layer is a temporal feature matrix containing 100 consecutive time steps (each time step corresponds to 1 second of data). Each time step covers four core features (Received Signal Strength Indicator, Time of Flight, Battery Voltage, and Ambient Temperature; here, we take four core features as an example). The first LSTM network contains 32 neurons and retains the temporal output (return_sequences=True), which can capture short-term burst features of the signal (such as electromagnetic interference caused by device start-up and shutdown). After the first layer, a dropout layer is connected with a dropout probability set to 0.2 to avoid model overfitting by randomly dropping some neuron connections. The second LSTM network also contains 32 neurons and outputs a one-dimensional feature vector, which can extract long-term trend features of the signal (such as gradual battery decay). Finally, a fully connected layer (Dense) outputs two prediction values, corresponding to the predicted value of the Received Signal Strength Indicator and the predicted value of the Time of Flight for the next time step, i.e., the measured value of RSSI and the predicted value of RSSI.

[0049] The advantages of using a two-layer recurrent neural network structure: Signal anomalies in the oil and gas environment have dual characteristics of short-term bursts (such as electromagnetic interference caused by equipment start-up and shutdown) and long-term trends (such as battery degradation). The two-layer network can capture short-term fluctuations and long-term trends respectively.

[0050] The basis for the input window length of 100 is: through experimental verification, 100 time steps (100 seconds) can cover 90% of the fault precursor cycle (such as the battery voltage drop trend usually lasting 50~150 seconds).

[0051] The training of the LSTM prediction model includes the following: For the dataset: 90 consecutive days of operational data from 30 nodes in an oil field were collected, with 70% being normal samples and 30% being fault samples (containing 12 types of faults). The training, validation, and test sets were divided in an 8:1:1 ratio. For the loss function: a combination of mean squared error (MSE) and mean absolute error (MAE) was used: the loss function Loss was 0.7 times MSE plus 0.3 times MAE, avoiding the sensitivity of MSE to extreme values ​​and improving the detection capability of small faults (e.g., small shifts in TOF). For the training parameters: the batch size was set to 32, the number of training epochs was set to 50, the learning rate was set to 0.001, and an early stopping strategy was adopted (the training stopped if the validation set loss did not decrease for 5 consecutive epochs). Finally, on the test set, the RSSI prediction error was no greater than 0.05, and the TOF prediction error was no greater than 0.2m.

[0052] Step S22: Determine the specific features and feature dimensions of the input LSTM prediction model based on the feature adaptive selection mechanism.

[0053] In step S22, the feature adaptive selection mechanism has been described in detail above.

[0054] Step S23: Output the RSSI prediction value and TOF prediction value for the next time step through the fully connected layer of the LSTM prediction model.

[0055] Step S23 above also includes the process of calculating outlier scores, specifically including: calculating the normalized error between the measured RSSI value and the predicted RSSI value, and the normalized error between the measured TOF value and the predicted TOF value, denoted as […]. and The normalized error between the measured and predicted RSSI values ​​is the absolute value of the difference between the measured and predicted RSSI values, divided by the difference between the maximum and minimum RSSI values. The normalized error between the measured and predicted TOF values ​​is the absolute value of the measured and predicted TOF values, divided by the difference between the maximum and minimum TOF values.

[0056] In this embodiment, a preliminary anomaly warning can be triggered when the normalization error between the measured RSSI value and the predicted RSSI value is greater than 0.3 and lasts for more than 5 time steps, or when the normalization error between the measured TOF value and the predicted TOF value is greater than 0.2.

[0057] In this embodiment, the core outputs of the LSTM model are the predicted value of the received signal strength indicator (RSSI) and the predicted value of the flight time (AST) for the next time step. These two outputs correspond to the two core diagnostic dimensions of communication link status and positioning accuracy status in the oil and gas scenario, respectively. The RSI directly reflects the working status of the communication module, and the AST directly reflects the working status of the positioning module. Together, they cover more than 80% of node fault scenarios (e.g., antenna attenuation in the communication module, crystal oscillator drift in the positioning module). The anomaly score calculation focuses on these two core outputs because: other six-dimensional features (e.g., voltage, temperature) are integrated into the model training process through a feature adaptive selection mechanism, and their impact is already reflected in the predicted values ​​(e.g., voltage attenuation leads to a decrease in the predicted value of the RSI); the subsequent rule engine verification stage (i.e., the process of using priority sorting and time window verification rule sets to determine the fault type) will introduce the measured data of all eight-dimensional features (e.g., temperature, voltage) to perform multi-dimensional verification of the anomaly score results, avoiding the limitations of single-dimensional judgment.

[0058] Step S3: Based on the output value, use priority sorting and time window verification rule set to determine the fault type.

[0059] The method of determining the fault type by using priority sorting and time window verification rule sets mainly includes the following sub-steps.

[0060] Step S31: First, sort the rules according to their preset priority, and then check the triggering conditions of each rule in order of priority from high to low. Step S32: For rules that meet the triggering conditions, verify whether the duration of the abnormal feature reaches a preset threshold. Step S33: When multiple rules of the same priority meet the triggering conditions, record the timestamp of the first time each rule meets the triggering conditions, and select the rule with the latest timestamp as the current diagnostic result.

[0061] In this technical solution, priority strategy and time window strategy are used to execute the set of rules.

[0062] More specifically, the above-mentioned method of using priority sorting and time window verification rule sets to determine the fault type includes the following:

[0063] Rule pre-sorting: All rules are sorted according to preset priority (high, medium, low). High priority rules (such as positioning module failure, signal link interruption) take precedence over medium priority rules (such as insufficient battery power), and medium priority rules take precedence over low priority rules (such as environmental interference warning).

[0064] Conditions are judged sequentially: the trigger conditions of each rule are checked in descending order of priority (for example, first check whether the trigger condition of the high-priority "location module failure" is met, and then check whether the trigger condition of the medium-priority "power module abnormality" is met).

[0065] Time window verification: For rules that meet the triggering conditions, it is necessary to further verify whether the "characteristic anomaly duration" has reached the preset threshold (e.g., high-priority rules need to last for 5 time steps, and medium-priority rules need to last for 8 time steps) to avoid false triggering caused by momentary interference.

[0066] Priority conflict resolution: When multiple rules with the same priority meet the triggering conditions at the same time, the "most recent triggering time" principle is used to determine the priority. Specifically, the timestamp of the first time each rule meets the triggering condition is recorded, the rule with the latest timestamp is selected as the current diagnostic result, and the full rule verification is restarted after 10 seconds to avoid missing potential compound faults.

[0067] The rules mentioned above include rule categories, triggering conditions, fault types, and priorities. All rules are stored to form a rule set.

[0068] For the rule set, this embodiment uses Table 1 as an example for demonstration: Table 1. Design of Partial Rule Sets .

[0069] Step S4: Classify the faults and execute alarms according to the constructed multi-level alarm mechanism. The multi-level alarm mechanism includes a three-level alarm system, which includes: When the fault type is a potential anomaly, a Level 1 early warning is triggered; when the fault is a confirmed fault, a Level 2 alarm is triggered; when the fault involves a safety threat, a Level 3 emergency shutdown is initiated, cutting off the power supply to the relevant equipment.

[0070] More specifically, for Level 1 warning alerts, all faults are potential anomalies (e.g., battery voltage drops from 3.7V to 3.5V and the LSTM prediction model predicts it will drop below 3.4V in the next 2 hours). Warning information is sent to the handheld terminal of the inspection personnel via LoRa communication, including the probability of the fault (e.g., 75% probability of battery degradation) and recommended measures (e.g., replace the battery within 12 hours).

[0071] For Level 2 alarms: the fault is a clear fault (such as signal link interruption), which triggers the on-site audible and visual alarm (sound intensity greater than or equal to 80dB) and uploads the fault log to the central control system via 5G. The log contains the original data sequence of the 10 minutes before the fault (which can be analyzed afterward).

[0072] For Level 3 emergency shutdown: all are faults involving safety threats (e.g., the positioning module cannot output position data for 1 minute continuously), the wellhead safety system will cut off the power to the relevant equipment and send an SMS to the safety supervisor (response time <500ms).

[0073] Based on the above, the solution in this embodiment can also generate a fault heatmap on the central control system interface, marking the location, type, and impact range of fault nodes. For example, when three nodes in a certain area simultaneously experience "signal link failure," it is automatically marked as "regional electromagnetic interference," and it is recommended to adjust the operating frequency of nearby frequency converters.

[0074] Existing technologies suffer from limitations such as single-indicator limitations, static threshold defects, lack of time-series prediction, and high computational complexity. Single-indicator limitations: analyzing only RSSI or TOF fails to distinguish between hardware faults and environmental interference (e.g., RSSI drop due to metal obstruction overlaps with antenna fault characteristics). Static threshold defects: fixed thresholds do not consider the impact of temperature and humidity on the signal (e.g., RSSI threshold needs to be dynamically adjusted by 3dBm for every 10°C increase in temperature). Lack of time-series prediction: only detecting data at the current moment cannot predict future faults (e.g., if battery voltage drops at a rate of 0.1V / day, an alarm will only be triggered after 3 days). High computational complexity: deep learning models (e.g., Bi-LSTM) have over 1 million parameters, making it impossible to deploy edge nodes (with no more than 256MB of memory).

[0075] In this embodiment, by integrating a multi-dimensional feature fusion diagnostic architecture, a lightweight LSTM edge computing model, a dynamic rule adaptive mechanism, and a three-level fault classification alarm system, the shortcomings of existing technologies are overcome, enabling rapid diagnosis and prediction of personnel positioning node faults, and ensuring real-time performance and security.

[0076] For the multi-dimensional feature fusion diagnostic architecture, an eight-dimensional feature vector space is constructed, including RSSI, TOF, battery voltage, temperature, operating mode, communication quality, signal multipath index, and node load rate. Through a dynamic weighting algorithm (the weights are automatically adjusted according to the environmental complexity), deep fusion of multi-dimensional features is achieved, which solves the problem that a single indicator cannot distinguish between environmental interference and hardware failure.

[0077] For the lightweight LSTM edge computing model, a two-layer LSTM network + Dropout regularization architecture is adopted. The number of model parameters is compressed to 12,000 through weight pruning (threshold 0.01) and 8-bit quantization technology. Combined with the LeakyReLU activation function (α=0.01), the convergence speed is improved by 25%. Ultra-real-time inference of 42ms / inference is achieved on the STM32H743 chip, which meets the response requirement of less than 100ms for oil and gas operations.

[0078] For the dynamic rule adaptive mechanism, an environmental parameter-diagnostic threshold mapping table (such as a temperature-RSSI threshold table) is established, and a rule self-updating algorithm is designed. By analyzing historical fault data, the rule parameters are automatically adjusted (such as adjusting the RSSI threshold every 5℃), so that the diagnostic rules can dynamically evolve with the environment.

[0079] For the three-level fault classification alarm system, a three-level response mechanism of early warning, alarm and emergency shutdown is divided based on the severity and development trend of the fault. A self-repair program for fault nodes and a redundant location algorithm for neighboring nodes are designed to build a closed-loop system of detection, early warning and handling, which reduces the average fault handling time from 120 seconds to 28 seconds.

[0080] Example 2 Based on Example 1, this example also proposes a rapid diagnosis system for personnel positioning node faults, combined with... Figure 1 It includes a data acquisition module, a preprocessing module, a predictive analysis module, and a decision execution module. The data acquisition module can perform data acquisition functions and collect data from multiple sources. The preprocessing module can preprocess the acquired multi-source data and finally generate multi-dimensional feature data. The predictive analysis module can use the trained LSTM prediction model to predict RSSI and TOF. The decision execution module can determine the fault type and execute the alarm.

[0081] In this embodiment, the acquisition module is connected to the preprocessing module, the preprocessing module is connected to the predictive analysis module, and the predictive analysis module is connected to the decision execution module. The acquisition module, preprocessing module, predictive analysis module, and decision execution module can work together to perform fault diagnosis and fault prediction of personnel positioning nodes.

Claims

1. A method for rapid diagnosis of personnel positioning node faults, characterized in that, Includes the following steps: S1, collect multi-source data and preprocess the multi-source data to generate multi-dimensional features; S2, input the multidimensional features into the trained LSTM prediction model according to the feature adaptive selection mechanism, and output the RSSI prediction value and the TOF prediction value. S3, based on the output value, uses priority sorting and time window verification rule set to determine the fault type; S4 classifies faults and executes alarms based on the constructed multi-level alarm mechanism.

2. The method for rapid diagnosis of personnel positioning node faults according to claim 1, characterized in that, Step S1 includes: acquiring several types of raw data respectively, performing analog-to-digital conversion through the ADC module, and transmitting the data directly to the MCU after the conversion is completed; then preprocessing the data, including outlier filtering, normalization processing and timing alignment processing.

3. A method for rapid diagnosis of personnel positioning node faults according to claim 1 or 2, characterized in that, The feature adaptive selection mechanism includes: A preset feature priority mapping table is used to divide features into core features, auxiliary features, and supplementary features based on the correlation between the fault type and features in the oil and gas scenario. When in basic diagnosis mode, only core features are enabled. When in precise diagnosis mode, auxiliary features are activated if the anomaly score of the core features is not less than 0.

2. Supplementary features are added when in difficult diagnosis mode.

4. The method for rapid diagnosis of personnel positioning node faults according to claim 3, characterized in that, Step S2 includes: S21, an LSTM prediction model is constructed based on a two-layer recurrent neural network structure, with the input layer being a temporal feature matrix; S22, determine the specific features and feature dimensions input to the LSTM prediction model based on the feature adaptive selection mechanism; S23, the fully connected layer of the LSTM prediction model outputs the RSSI prediction value and TOF prediction value for the next time step.

5. The method for rapid diagnosis of personnel positioning node faults according to claim 4, characterized in that, Step S23 further includes calculating the anomaly score, specifically: calculating the normalized error between the measured RSSI value and the predicted RSSI value, and the normalized error between the measured TOF value and the predicted TOF value, denoted as […]. and .

6. A method for rapid diagnosis of personnel positioning node faults according to claim 1, 2, or 4, characterized in that, Using priority sorting and time window verification rule sets to determine fault type includes: The rules are categorized and sorted according to preset priority, and the triggering conditions of each rule are verified in descending order of priority. For rules that meet the triggering conditions, verify whether the duration of the abnormal feature reaches a preset threshold; When multiple rules of the same priority meet the triggering condition, record the timestamp of the first time each rule meets the triggering condition, and select the rule with the latest timestamp as the current diagnostic result.

7. The method for rapid diagnosis of personnel positioning node faults according to claim 6, characterized in that, The rules include rule categories, triggering conditions, fault types, and priorities. All rules are stored to form a rule set.

8. A method for rapid diagnosis of personnel positioning node faults according to claim 1, 2, or 4, characterized in that, The multi-level alarm mechanism includes a three-level alarm system, which includes: When the fault type is a potential anomaly, a Level 1 early warning is triggered; when the fault is a confirmed fault, a Level 2 alarm is triggered; when the fault involves a safety threat, a Level 3 emergency shutdown is initiated, cutting off the power supply to the relevant equipment.

9. The method for rapid diagnosis of personnel positioning node faults according to claim 3, characterized in that, The core feature is a four-dimensional feature, the auxiliary feature is a two-dimensional feature, and the supplementary feature is a two-dimensional feature.

10. A rapid fault diagnosis system for personnel positioning nodes, applicable to the rapid fault diagnosis method for personnel positioning nodes as described in any one of claims 1-9, characterized in that, include: The data acquisition module collects data from multiple sources. The preprocessing module performs preprocessing on multi-source data; The predictive analysis module uses an LSTM prediction model for prediction. The decision execution module includes fault type identification and alarm execution.

Images