Multi-sensing threshold self-learning method for a flapover berth device

Abstract

The application discloses a kind of multi-sensing threshold self-learning methods of plate type berth equipment, and the application relates to the technical field of mechanical equipment state monitoring and fault diagnosis, comprising the following steps: S1: collecting the multi-source heterogeneous data of plate type berth equipment at multiple preset monitoring points, the multi-source heterogeneous data at least includes electrical parameter data, mechanical vibration data, position angle data and environmental compensation data.The multi-sensing threshold self-learning method of the plate type berth equipment, realizes the fine identification and dynamic threshold generation of equipment operating state.Can effectively distinguish the parameter drift caused by environmental temperature change, lubrication state fluctuation or mechanical natural aging and real fault characteristics, avoid fixed threshold after seasonal change or equipment wear frequently trigger false alarm, ensure that sudden mechanical damage or electrical anomaly is captured in time, improve the accuracy of fault detection and the continuity of equipment operation.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment condition monitoring and fault diagnosis technology, specifically a multi-sensor threshold self-learning method for flip-type berth equipment. Background Technology

[0002] Flip-over parking systems, such as common automated parking garage platforms, flip-over boarding bridges, or cargo loading / unloading platforms, use a flipping mechanism to park vehicles or transfer goods. They are widely used in logistics, warehousing, and transportation hubs. These systems typically consist of a drive unit, transmission mechanism, load-bearing flip-over platform, and control system. Their operational stability directly affects operational safety and efficiency.

[0003] Currently, to ensure the safe operation of tilting berth equipment, multiple sensors are typically used for status monitoring. These include angle sensors to monitor the tilting plate position, current / voltage sensors to monitor motor load, and proximity switches to monitor positioning. The control system compares the data collected by these sensors with preset fixed alarm thresholds. If the monitored value exceeds the threshold range, a fault or safety hazard is detected, triggering an alarm. Publication number "CN114035555A" describes a "PLC controller fault detection system," which collects multi-source signals such as voltage, current, temperature, and sound waves through a data acquisition system. The system then uses data threshold and comparison modules in the analysis system to compare real-time data with preset fixed thresholds to determine the cause of the fault, achieving online real-time monitoring of the controller's operating status. This solution has certain advantages in fault detection for single devices or under stable operating conditions.

[0004] However, the aforementioned patents and existing fault detection methods for flip-type berth equipment still have the following shortcomings: First, the alarm thresholds they rely on are usually fixed values ​​set at the factory or human experience values, which are static thresholds.

[0005] In practical applications, tilting berth equipment operates in open or semi-open environments for extended periods. Its mechanical structure experiences slow, non-fault-related drift in operating parameters due to factors such as wear, changes in lubrication, temperature fluctuations, and foreign object obstruction. Fixed thresholds cannot adapt to these long-term, gradual environmental and mechanical characteristic changes, easily leading to false alarms or missed alarms. Secondly, for different batches and varying degrees of wear of the same model of equipment, the range of normal operating parameters also varies individually, making it difficult to achieve accurate, personalized monitoring using uniform fixed thresholds. Finally, when equipment frequently triggers false alarms due to environmental changes, manual intervention to adjust the thresholds is required, which is not only time-consuming and labor-intensive but also relies heavily on the experience of maintenance personnel, indicating a low level of automation. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-sensor threshold self-learning method for flip-type berth equipment to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a multi-sensor threshold self-learning method for a flip-type berth device, comprising the following steps:

[0008] S1: Collect multi-source heterogeneous data of the flip-type berth equipment at multiple preset monitoring points. The multi-source heterogeneous data includes at least electrical parameter data, mechanical vibration data, position angle data, and environmental compensation data.

[0009] S2: Perform time-series segmentation on the collected multi-source heterogeneous data, dividing it into several event segments with physical meaning according to the equipment operation stage;

[0010] S3: Each event segment is simultaneously input into the first and second inspection units for parallel analysis; the first inspection unit performs dynamic envelope analysis on the event segment based on the statistical process control model to generate a first inspection result; the second inspection unit performs similarity matching between the event segment and the preset standard working condition template based on the working condition matching model to generate a second inspection result.

[0011] S4: Input the first and second test results into the arbitration verification unit. The arbitration verification unit performs no less than two cross-validations and iterative optimizations on the two test results according to the preset arbitration logic to generate an intermediate judgment result.

[0012] S5: Input the intermediate judgment results into the health assessment model for third-level verification. The health assessment model generates the final diagnosis result based on mechanical wear trend analysis.

[0013] S6: If the final diagnosis result is a deviation from normal operating conditions, the statistical baseline parameters of the first inspection unit and / or the standard operating condition template of the second inspection unit are dynamically updated based on the result; if the final diagnosis result is a fault state, an alarm signal is triggered and the fault characteristic data is locked.

[0014] Preferably, the electrical parameter data in the multi-source heterogeneous data includes the three-phase current and voltage values ​​of the drive motor; the mechanical vibration data includes the vibration spectrum characteristic values ​​of key bearing parts; the position angle data includes the real-time change curve of the flap rotation angle; and the environmental compensation data includes at least the ambient temperature value and lubrication state parameters, wherein the lubrication state parameters are obtained by monitoring the oil film thickness of key transmission components using an ultrasonic sensor. The electrical parameter data is collected synchronously through current transformers and voltage transformers, the vibration spectrum characteristic values ​​are extracted from the vibration sensor signals after Fourier transform, and the oil film thickness is calculated based on the propagation time and attenuation degree of ultrasonic waves in the oil film.

[0015] Preferably, the specific method for dividing the equipment operation phase into several physically meaningful event segments is as follows: Using a sliding time window technique, based on the start / stop signals of the drive motor and the rate of change of the tilting plate's angular velocity, the continuous data stream is sequentially divided into a start-up impact segment, a uniform speed tilting segment, a deceleration segment upon reaching the destination segment, and a stationary standby segment. A timestamp and operating condition label are added to each segment. The start and end times of each event segment are precisely determined based on the motor's operating state and the changing characteristics of the tilting plate's angular velocity. The timestamp contains the start and end time information of the segment, and the operating condition label is uniquely marked according to the segment's operating phase attributes.

[0016] Preferably, the specific process of the first inspection unit performing dynamic envelope analysis is as follows: Statistical analysis is performed on event segments under the same operating condition label whose historical operation count reaches a preset number; the mean and standard deviation of each characteristic parameter are calculated; the mean plus or minus n times the standard deviation is used as the upper and lower limits of the dynamic envelope; the value of n is adaptively adjusted according to the number of equipment operations. The characteristic parameters analyzed include the peak current of the drive motor, the amplitude of the characteristic frequency of the vibration spectrum, and the average angular velocity of the flapper, etc. The value of n is adjusted accordingly based on different operating stages of the equipment, such as the break-in period, the stable period, and the wear period.

[0017] Preferably, the operating condition matching model in the second verification unit is an autoencoder structure built based on a long short-term memory network. The specific method for similarity matching is as follows: the current event segment is input into the autoencoder to calculate the reconstruction error value. If the reconstruction error value is less than a preset threshold, it is determined to match the standard operating condition template; otherwise, it is determined to be a template mismatch. The standard operating condition template includes a preset general operating condition template and a self-learned, unique operating condition template. The unique operating condition template is generated by clustering event segments marked as special operating conditions in historical operating data. The reconstruction error value of the autoencoder is calculated using root mean square error or mean absolute error. The clustering operation of the unique operating condition template is initiated after a preset number of special operating condition segments under the same operating condition label have accumulated.

[0018] Preferably, the cross-validation and iterative optimization process performed by the arbitration verification unit includes:

[0019] First round of arbitration: If the first inspection result is determined to be out of tolerance while the second inspection result is determined to be in compliance with the working condition template, the arbitration unit will not generate an alarm command, but will mark the current event segment as special working condition data and output it to the standard working condition template library for template update;

[0020] If the first test result is normal but the second test result is template mismatch, the arbitration unit triggers a mechanical characteristic drift check instruction, packages the current event fragment and its test result, and sends it to the health assessment model.

[0021] If both the first and second inspection results are deemed abnormal, the arbitration unit directly generates a suspected fault signal and proceeds to the second round of verification. The mechanical characteristic drift check command packages and sends the complete data packet of the event segment along with the inspection data of each characteristic parameter. The suspected fault signal triggers the health assessment model to initiate a fast verification channel for priority processing.

[0022] Preferably, the health assessment model is constructed based on the Weibull distribution. The specific method for the third verification is as follows: receiving event fragment data sent by the arbitration unit, extracting characteristic parameters related to mechanical wear, and calculating the deviation between the current wear level and the historical wear trend curve; if the deviation is within a preset gradual range, it is determined to be normal aging, and the event fragment is fed back to the first verification unit as new baseline data for statistical parameter updates; if the deviation exceeds the gradual range and exhibits a step characteristic, it is determined to be a fault state by combining the results of the first and second verifications. Wear-related characteristic parameters include the peak value of the drive motor starting current, the effective value of the uniform speed operating current, and the thickness value of the lubricating grease film, etc., and the deviation is quantified using residual or relative deviation methods.

[0023] Preferably, the specific method for dynamically updating the statistical baseline parameters of the first-level inspection unit is as follows: new event segments judged by the health assessment model as normal aging are included in the historical dataset corresponding to the operating condition label, and the mean and standard deviation are recalculated, so that the dynamic envelope adaptively drifts with the degree of equipment aging. The specific method for dynamically updating the standard operating condition template of the second-level inspection unit is as follows: event segments marked as special operating conditions by the arbitration unit are clustered to generate new operating condition templates, or the feature weights of the original operating condition templates are adjusted. The historical dataset of the first-level inspection unit uses a first-in-first-out queue structure to maintain a fixed capacity of sample data, and the clustering algorithm of the second-level inspection unit selects either density clustering or mean-shift clustering.

[0024] Preferably, the specific method for triggering the alarm signal and locking the fault feature data is as follows: the fault type corresponding to the final diagnostic result, the event fragment data at the time of the fault occurrence, and the test results that led to the fault determination are associated and stored in the fault database. Simultaneously, the alarm information is sent to the backend terminal via the communication module, and the fault feature data is recorded for rapid matching of similar faults in the future. The fault database assigns a unique fault number to each fault feature data and establishes a multi-level index based on the fault type. The alarm information is pushed to the monitoring interface of the backend terminal in real time via the communication module.

[0025] Preferably, the method is also applicable to fault detection of other mechanical equipment with tilting mechanisms, including vehicle loading platforms in automated parking garages, tilting boarding bridges, and cargo loading and unloading platforms. Its application involves mapping the corresponding equipment's operational characteristic parameters to the data acquisition dimensions of the multi-source heterogeneous data, and adapting the model according to the equipment's standard operating condition template. The start and end rules for event segment segmentation of different tilting mechanical equipment will be redefined based on their own operational stages, and the Weibull distribution parameters of the health assessment model will be recalibrated in conjunction with the design life and wear mechanism of the equipment's mechanical components.

[0026] This invention provides a multi-sensor threshold self-learning method for flip-type berth equipment. It has the following beneficial effects:

[0027] This method constructs a multi-level parallel analysis architecture consisting of a first-level verification unit, a second-level verification unit, and a health assessment model. It also introduces an arbitration verification unit to perform at least two cross-validations and iterative optimizations, achieving refined identification of equipment operating status and dynamic threshold generation. This method effectively distinguishes parameter drift caused by changes in ambient temperature, lubrication fluctuations, or natural mechanical aging from actual fault characteristics. It avoids frequent false alarms triggered by fixed thresholds due to seasonal changes or equipment wear, while ensuring timely detection of sudden mechanical damage or electrical anomalies, thereby improving the accuracy of fault detection and the continuity of equipment operation.

[0028] This method feeds back special operating condition data marked by arbitration verification units to a standard operating condition template library for clustering and updating. It also uses normal aging data determined by the health assessment model to dynamically adjust the statistical baseline of the statistical process control model, enabling the threshold generation mechanism to have continuous self-learning and adaptive capabilities, reducing the workload of manual threshold adjustment. Furthermore, through the associated storage and rapid matching of fault feature data, this method achieves rapid identification and experience reuse of similar faults. It can also be extended to various tipping machinery such as automated parking garage platforms, flip-up boarding bridges, and freight loading / unloading platforms through parameter mapping and model adaptation, enhancing the versatility and engineering application value of this detection method. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a multi-sensor threshold self-learning method for a flip-type berth device according to the present invention.

[0030] Figure 2 This is a data flow diagram of the modules of the multi-sensor threshold self-learning method for a flip-type berth device according to the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0032] Please see Figure 1 and Figure 2 This invention provides a technical solution: a multi-sensor threshold self-learning method for flip-type berth equipment, comprising the following steps:

[0033] S1: Collect multi-source heterogeneous data of the flip-type berth equipment at multiple preset monitoring points. The multi-source heterogeneous data includes at least electrical parameter data, mechanical vibration data, position angle data, and environmental compensation data.

[0034] S2: Perform time-series segmentation on the collected multi-source heterogeneous data, dividing it into several event segments with physical meaning according to the equipment operation stage;

[0035] S3: Input each event segment into the first and second inspection units simultaneously for parallel analysis; the first inspection unit performs dynamic envelope analysis on the event segment based on the statistical process control model to generate the first inspection result; the second inspection unit performs similarity matching between the event segment and the preset standard working condition template based on the working condition matching model to generate the second inspection result.

[0036] S4: Input the first and second test results into the arbitration verification unit. The arbitration verification unit performs no less than two cross-validations and iterative optimizations on the two test results according to the preset arbitration logic to generate intermediate judgment results.

[0037] S5: Input the intermediate judgment results into the health assessment model for third-level verification. The health assessment model generates the final diagnosis result based on mechanical wear trend analysis.

[0038] S6: If the final diagnosis result is a deviation from normal operating conditions, the statistical baseline parameters of the first inspection unit and / or the standard operating condition template of the second inspection unit will be dynamically updated based on the result; if the final diagnosis result is a fault state, an alarm signal will be triggered and the fault characteristic data will be locked.

[0039] It should be further explained that, in the specific implementation process, the multi-sensor threshold self-learning method of this flap berth equipment first deploys multiple types of sensors at the locations of the equipment's drive motor, key transmission bearings, flap shaft, and hydraulic lubrication system to collect in real time electrical parameter data, including the three-phase current and voltage values ​​of the drive motor, mechanical vibration data, including the bearing vibration spectrum characteristic values, position angle data, including the real-time change curve of the flap rotation angle, and environmental compensation data, including ambient temperature values ​​and lubrication state parameters obtained by monitoring oil film thickness through ultrasonic sensors.

[0040] After receiving the aforementioned multi-source heterogeneous data in a time sequence, the central control system uses sliding time window technology to automatically divide the continuous data stream into start-up impact segments, uniform speed flipping segments, arrival deceleration segments, and stationary standby segments based on the start-stop signals of the drive motor and the rate of change of the flipping angular velocity. It also adds corresponding timestamps and operating condition tags to each segment, thereby converting the raw data into event segments with physical meaning.

[0041] Subsequently, each event segment is simultaneously sent to two parallel testing units for the first round of analysis. The first testing unit performs dynamic envelope analysis on the event segment based on a statistical process control model. Specifically, it statistically analyzes event segments that have reached a preset number of historical runs under the same operating condition label, calculates the mean and standard deviation of each characteristic parameter, and generates the upper and lower limits of the dynamic envelope by adding or subtracting n times the standard deviation from the mean. The value of n is adaptively adjusted according to the cumulative number of runs of the equipment, thereby outputting the first test result.

[0042] The second verification unit is based on an autoencoder structure built from a long short-term memory network. It performs similarity matching between the event fragment and a preset standard operating condition template. It calculates the reconstruction error value after the input fragment is encoded and then decoded, and compares this value with a preset threshold. If the reconstruction error is less than the threshold, it is determined to match the standard operating condition template; otherwise, it is determined to be a template mismatch, and thus outputs the second verification result.

[0043] The first and second test results are then input into the arbitration verification unit. This unit performs no less than two cross-validations and iterative optimizations based on the preset arbitration logic. Specifically, in the first round of arbitration, if the first test result determines that the data is out of tolerance while the second test result determines that the data conforms to the operating condition template, the arbitration unit does not directly generate an alarm command. Instead, it marks the event fragment as special operating condition data and outputs it to the standard operating condition template library for subsequent template clustering and updating.

[0044] If the first test result is normal but the second test result is template mismatch, the arbitration unit triggers a mechanical characteristic drift check command, packages the event fragment and its test result, and sends it to the health assessment model.

[0045] If both the first and second inspection results are deemed abnormal, the arbitration unit will directly generate a suspected fault signal and proceed to the second round of verification.

[0046] The health assessment model is built on the Weibull distribution. After receiving the data sent by the arbitration unit, it extracts the feature parameters related to mechanical wear and calculates the deviation between the current wear level value and the historical wear trend curve. When the deviation is within the preset gradual range, it is determined to be a normal aging process. At this time, the event segment is used as the new baseline data and fed back to the first verification unit. The mean and standard deviation are recalculated in the historical data of the corresponding working condition label, so that the dynamic envelope can adaptively drift with the aging degree of the equipment.

[0047] When the deviation exceeds the gradual range and exhibits a step change characteristic, the system combines the first and second test results to determine a fault state. At this time, the system triggers an alarm signal, associates and stores the fault type, event fragment data at the time of the fault occurrence, and the test results that led to the fault determination in the fault database. At the same time, the alarm information is sent to the back-end terminal through the communication module, and the fault characteristic data is recorded for rapid matching of similar faults in the future.

[0048] If the special working condition data marked by the arbitration unit reaches the clustering condition after a period of accumulation, the system will generate a new working condition template or adjust the feature weights of the original working condition template, thereby realizing the dynamic updating of the standard working condition template library.

[0049] The electrical parameter data in the multi-source heterogeneous data includes the three-phase current and voltage values ​​of the drive motor; the mechanical vibration data includes the vibration spectrum characteristics of key bearing parts; the position angle data includes the real-time change curve of the flap rotation angle; the environmental compensation data includes at least the ambient temperature value and lubrication status parameters, which are obtained by monitoring the oil film thickness of key transmission components using ultrasonic sensors.

[0050] It should be further explained that, in the specific implementation process, the method first installs current transformers and voltage transformers at the ends of the three-phase windings of the drive motor of the flip-type berth equipment, respectively, to collect the three-phase current and voltage values ​​of the drive motor in real time. The sampling frequency is set to an integer multiple of the equipment control cycle to synchronize the motor operation status.

[0051] An acceleration vibration sensor is installed at the critical transmission bearing housing of the equipment. The vibration signal collected by the sensor is subjected to Fourier transform to extract vibration spectrum feature values, including the amplitude of the bearing's characteristic frequency, which are used to characterize the operational stability of the mechanical components.

[0052] An absolute encoder is installed at the rotation axis of the flap. The encoder continuously outputs angle pulse signals as the flap rotates. The control system calculates the real-time angle position of the flap at any time based on the cumulative pulse value and records the angle change curve throughout the entire process from start to finish.

[0053] A temperature sensor is installed in the environmental monitoring unit of the equipment. The temperature sensor is installed close to the drive motor and collects the ambient temperature value in real time. At the same time, an ultrasonic sensor is installed near the grease injection port of the flip plate shaft and bearing. The ultrasonic sensor emits ultrasonic waves to the grease film and receives the reflected waves. The specific value of the current oil film thickness is calculated based on the propagation time and attenuation degree of the ultrasonic waves in the oil film as the lubrication status parameter.

[0054] All the electrical parameter data, mechanical vibration data, position and angle data, and environmental compensation data collected above are synchronously packaged and uploaded to the central control system via the data acquisition card, serving as the basic input data for subsequent event segmentation and multi-model parallel analysis.

[0055] The specific method for dividing the equipment into several physically meaningful event segments based on the equipment's operating phase is as follows: using sliding time window technology, based on the start / stop signal of the drive motor and the rate of change of the flip plate's angular velocity, the continuous data stream is sequentially divided into start-up impact segment, uniform speed flipping segment, arrival deceleration segment, and stationary standby segment, and timestamps and operating condition tags are added to each segment.

[0056] It should be further explained that, in the specific implementation process, this method uses the central control system to monitor the start and stop signals of the drive motor and the rate of change of the angular velocity of the flapper in real time. When the control system detects that the motor start signal jumps from low level to high level and the angular velocity of the flapper increases positively from zero within a preset time window, it determines that the equipment has entered the start-up stage. At this time, the moment before the angular velocity begins to increase is taken as the starting point of the start-up impact segment, and the moment when the angular velocity first reaches and stabilizes at 90% of the target angular velocity is taken as the ending point of the start-up impact segment. Thus, the start-up impact segment containing the peak value of the start-up current and the characteristics of the impact vibration is cut out.

[0057] Subsequently, when the angular velocity of the flip plate fluctuates around the target angular velocity and the fluctuation amplitude does not exceed the preset threshold, the control system determines this continuous process as the uniform speed flipping stage and cuts the continuous data segment from the end point of the starting impact segment to the moment when the angular velocity begins to continuously decrease into a uniform speed flipping segment.

[0058] When the control system detects that the flap is approaching the target position, triggers the limit switch signal, or the motor control command switches to deceleration mode, it determines that the equipment has entered the deceleration phase. The continuous data segment from the moment when the angular velocity begins to decrease continuously until the angular velocity decreases to zero and the flap reaches the target angle is cut into the deceleration segment.

[0059] During the period when the flapping angular velocity remains at zero and the motor has no drive signal, the control system divides the data of this period into static standby segments.

[0060] For each event segment that has been cut as described above, the control system automatically adds a corresponding start time timestamp, end time timestamp, and operating condition label. The operating condition label is marked as starting impact, uniform speed flipping, deceleration in place, or static standby according to the stage attributes of the segment. The labeled segment data is stored in the historical database for subsequent dynamic envelope calculation of the statistical process control model and template comparison and updating of the operating condition matching model.

[0061] The specific process of the first verification unit performing dynamic envelope analysis is as follows: statistical analysis is performed on event segments under the same working condition label whose historical running number reaches a preset number, the mean and standard deviation of each characteristic parameter are calculated, and the mean plus or minus n times the standard deviation is used as the upper and lower limits of the dynamic envelope, and the value of n is adaptively adjusted according to the number of times the equipment runs.

[0062] It should be further explained that, in the specific implementation process, the first verification unit reads all historical event segments with the same working condition label as the current event segment from the historical database. When the number of historical runs accumulated under the same working condition label reaches the preset minimum number threshold, the various characteristic parameters under the working condition label are statistically analyzed. These characteristic parameters include at least the peak and effective values ​​of the three-phase current of the drive motor, the amplitude of the vibration spectrum at the characteristic frequency, the average angular velocity of the flapper in the corresponding stage, and the overshoot angle when it is in place.

[0063] For each feature parameter, the first test unit calculates its arithmetic mean and sample standard deviation in the historical dataset. Specifically, the calculation method is to sum the values ​​of the feature parameter in all historical segments under the same working condition label, divide by the number of historical segments to obtain the mean, and then take the square root of the sum of squares of the deviations of each value from the mean, divided by the number of segments minus one, to obtain the standard deviation.

[0064] Subsequently, the upper and lower limits of the dynamic envelope of this feature parameter are generated by adding or subtracting n times the standard deviation from the mean. The determination of the value of n is related to the cumulative number of times the equipment has been run. When the cumulative number of times the equipment has been run is within the break-in period, the value of n is small to tighten the envelope. When the cumulative number of times the equipment has been run enters the stable period, n returns to the baseline value. When the cumulative number of times the equipment has been run reaches the wear period threshold, the value of n is appropriately increased to adapt to the parameter drift caused by equipment aging.

[0065] Once the dynamic envelope is generated, it serves as the normal fluctuation range of the characteristic parameters corresponding to the current event segment. It is used to compare with the actual characteristic value of the current segment. If the actual characteristic value is within the upper and lower limits of the envelope, the first test result is judged as normal. If it exceeds the upper and lower limits, it is judged as out of tolerance.

[0066] Meanwhile, after each judgment is completed, the feature data of the current event segment will be included in the historical dataset based on the feedback results of the subsequent arbitration verification unit or health assessment model, so as to be used for the recalculation of statistical parameters in the next round, thereby realizing the continuous updating of the dynamic envelope according to the actual operating status of the equipment.

[0067] The operating condition matching model in the second verification unit is an autoencoder structure built on a long short-term memory network. The specific method of similarity matching is as follows: the current event segment is input into the autoencoder to calculate the reconstruction error value. If the reconstruction error value is less than the preset threshold, it is determined to match the standard operating condition template; otherwise, it is determined to be a template mismatch. The standard operating condition template includes a preset general operating condition template and a self-learned specific operating condition template. The specific operating condition template is generated by clustering event segments marked as special operating conditions in historical operating data.

[0068] It should be further explained that, in the specific implementation process, the second verification unit is equipped with an autoencoder model based on a long short-term memory network. Before being put into online use, this model is first trained using multiple event segments collected during the normal operation of the device in the past. The training process adopts an unsupervised learning method, taking each event segment as an input sequence, compressing it into a low-dimensional latent space vector layer by layer through the encoder network, and then reconstructing the output sequence with the same dimension as the input sequence from the latent space vector through the decoder network. The goal of model training is to minimize the reconstruction error between the input sequence and the output sequence. After training, the network weight parameters of the encoder and decoder are fixed.

[0069] During the online matching phase, when a new current event segment is input into the trained autoencoder, the model automatically calculates the reconstruction error value between the segment after encoding and decoding and the original input. The reconstruction error value is either the root mean square error or the mean absolute error. The preset threshold is determined by extracting the error distribution statistics from the reconstruction error values ​​of all event segments in the training set and setting the upper limit value or a specific quantile value of the statistics as the preset threshold.

[0070] If the reconstruction error value of the current event segment is less than the preset threshold, it is determined that the current event segment matches the general pattern in the standard working condition template library, and the second test result is output as a normal match. If the reconstruction error value is greater than or equal to the preset threshold, it is determined that the template is mismatched, and the second test result is output as an abnormal match.

[0071] The standard operating condition template library consists of two parts. The first part is the preset general operating condition template, which is pre-built and fixed in the system based on the equipment design parameters and factory test data, covering the typical operating modes of the equipment under standard environment and standard load.

[0072] The second part is the self-learning generated unique operating condition template. The generation process of this template is as follows: when the arbitration verification unit marks a certain event segment as special operating condition data, the system stores the segment in the clustering buffer. When the number of special operating condition segments under the same operating condition label in the buffer reaches a preset number, the system performs feature extraction and unsupervised clustering on the batch of segments. After clustering, the common feature patterns of segments in the same category are extracted and a new unique operating condition template is generated. Alternatively, the weight coefficients of each feature dimension in the original similar operating condition template are adaptively adjusted according to the clustering results, so that the template can more accurately represent the actual operating status of the equipment under a specific environment or specific load.

[0073] The cross-validation and iterative optimization process performed by the arbitration verification unit includes:

[0074] First round of arbitration: If the first inspection result is determined to be out of tolerance while the second inspection result is determined to be in compliance with the working condition template, the arbitration unit will not generate an alarm command, but will mark the current event segment as special working condition data and output it to the standard working condition template library for template update;

[0075] If the first test result is normal but the second test result is template mismatch, the arbitration unit triggers a mechanical characteristic drift check instruction, packages the current event fragment and its test result, and sends it to the health assessment model.

[0076] If both the first and second inspection results are deemed abnormal, the arbitration unit will directly generate a suspected fault signal and proceed to the second round of verification.

[0077] It should be further explained that, in the specific implementation process, the arbitration verification unit continuously receives the first verification result from the first verification unit and the second verification result from the second verification unit. The arbitration verification unit has three sets of arbitration logic branches preset inside.

[0078] When the first test result determines that the feature parameters of the current event segment exceed the range of the dynamic envelope, and the second test result determines that the segment matches the standard working condition template normally, the arbitration verification unit executes the first logic branch, that is, without triggering any alarm command, it marks the event segment as special working condition data and outputs the segment together with its timestamp and working condition label to the clustering buffer of the standard working condition template library for subsequent clustering generation of special working condition templates or weight adjustment of existing templates.

[0079] When the first test result determines that the current event segment is within the range of the dynamic envelope, while the second test result determines that the segment does not match the standard working condition template, the arbitration verification unit executes the second logic branch, that is, triggers the mechanical characteristic drift check instruction. This instruction packages the data packet of the current event segment, the actual values ​​of each characteristic parameter of the segment in the first test result and their relative position information with the dynamic envelope, and the specific value of the reconstruction error in the second test result together to generate a data packet to be verified, and sends the data packet to the input queue of the health assessment model.

[0080] When both the first and second inspection results are determined to be abnormal, the arbitration verification unit executes the third logic branch, that is, directly generates a suspected fault signal carrying a double abnormal label, and temporarily stores the suspected fault signal along with the data packet of the current event fragment in the high-priority verification buffer area, waiting to enter the second round of verification process. At the same time, the suspected fault signal triggers the health assessment model to start the fast verification channel.

[0081] The health assessment model is based on the Weibull distribution. The specific method of its third verification is as follows: receiving event fragment data sent by the arbitration unit, extracting characteristic parameters related to mechanical wear, and calculating the deviation between the current wear level value and the historical wear trend curve; if the deviation is within the preset gradual range, it is judged as normal aging and the event fragment is fed back to the first verification unit as new baseline data for statistical parameter update; if the deviation exceeds the gradual range and shows a step characteristic, it is judged as a fault state by combining the results of the first verification and the second verification.

[0082] It should be further explained that, in the specific implementation process, the health assessment model is deployed with a life prediction engine based on the Weibull distribution. In the early stage of equipment operation, the engine initializes a set of shape parameters and scale parameters based on the design life and historical failure data of similar equipment. As the actual operating time of the equipment accumulates, the model continuously receives and records the characteristic parameters related to mechanical wear in each event segment. These characteristic parameters include at least the peak value of the starting current of the drive motor, the effective value of the uniform speed operating current, the total vibration energy value during the deceleration phase, and the grease film thickness value monitored by the ultrasonic sensor.

[0083] When the data packet to be verified sent by the arbitration verification unit arrives at the input queue of the health assessment model, the model first parses the wear-related characteristic parameters of the current event segment from the data packet. At the same time, it retrieves the historical sequence values ​​of the same characteristic parameters from previous event segments with the same operating condition label as the event segment from the historical database. Based on these historical sequence values, it fits the wear trend curve under the current operating condition. The curve uses the cumulative number of times the equipment has been run or the cumulative running time as the horizontal axis and the characteristic parameter value as the vertical axis, reflecting the natural drift law of the characteristic parameter as the equipment ages.

[0084] The model then calculates the degree of deviation between the current feature parameter value and the predicted value of the wear trend curve at the corresponding horizontal coordinate position. This degree of deviation is quantified using either residual or relative deviation.

[0085] The model has a preset gradient range threshold, which is determined based on the failure probability interval corresponding to different confidence levels in the Weibull distribution. When the calculated deviation is within the gradient range threshold, the health assessment model determines that the change in the current characteristic parameter belongs to the random fluctuation in the normal aging process of the equipment. At this time, the current event segment is marked as normal aging data, and all the characteristic parameters of the segment are fed back to the first verification unit as new sample points for the expansion of the historical dataset under the working condition label and the recalculation of the dynamic envelope statistical parameters. At the same time, the judgment result is sent back to the arbitration verification unit to close the current verification process.

[0086] When the calculated deviation exceeds the gradual range threshold and the deviation exhibits a monotonically increasing or step-like jump characteristic in multiple consecutive event segments, the health assessment model determines that the equipment has abnormal wear or sudden mechanical damage. At this time, the determination result is logically ANDed with the first test result and the second test result. Only when all three test results point to an abnormal state is the final comprehensive determination a fault state made. The fault state, along with the specific characteristic parameter changes that led to the determination, the local inflection point information of the wear trend curve, and the specific value of the deviation, are encapsulated together as the final diagnostic result output.

[0087] The specific method for dynamically updating the statistical baseline parameters of the first-level inspection unit is as follows: the feature values ​​of new event segments that are judged as normal aging by the health assessment model are included in the historical dataset of the corresponding operating condition label, and the mean and standard deviation are recalculated so that the dynamic envelope adaptively drifts with the degree of equipment aging. The specific method for dynamically updating the standard operating condition template of the second-level inspection unit is as follows: the event segments marked as special operating conditions by the arbitration unit are clustered to generate new operating condition templates, or the feature weights of the original operating condition templates are adjusted.

[0088] It should be further explained that, in the specific implementation process, the statistical baseline parameter update mechanism of the first inspection unit is directly related to the judgment output of the health assessment model. When the health assessment model determines that a certain event segment is normal aging data, all feature parameter values ​​of the segment are automatically pushed to the historical dataset storage area of ​​the corresponding working condition label in the first inspection unit. This storage area uses a first-in-first-out queue structure to maintain a fixed capacity of historical samples. When a new sample enters the queue, the oldest sample at the head of the queue is automatically removed to maintain the timeliness of the dataset.

[0089] Subsequently, the first verification unit re-executes statistical process control analysis on the updated complete historical dataset under the operating condition label, that is, recalculates the arithmetic mean and sample standard deviation of each feature parameter, and regenerates the upper and lower limits of the dynamic envelope based on the updated mean and standard deviation, so as to achieve smooth drift of the statistical baseline as the equipment ages.

[0090] The standard operating condition template update mechanism of the second verification unit is directly related to the label output of the arbitration verification unit. When the arbitration verification unit marks a certain event segment as special operating condition data, the segment is stored in the temporary storage area of ​​the corresponding operating condition label in the clustering buffer. The system monitors the number of segments under the same operating condition label in the buffer in real time. When the number reaches the preset clustering start threshold, the system calls the clustering analysis program to perform feature extraction and unsupervised clustering on all special operating condition segments in the buffer. The clustering algorithm adopts one of density clustering or mean-shift clustering, and automatically divides the clusters according to the density of the sample distribution in the feature space.

[0091] The center vector of each cluster is extracted as the typical feature of the corresponding working condition mode. The system compares the similarity of this typical feature with the templates in the existing standard working condition template library. If the similarity with the existing template is lower than the preset merging threshold, it is determined that a new working condition mode has appeared. At this time, the typical feature is defined as a new unique working condition template and stored in the template library. If the similarity with an existing template is higher than the preset merging threshold, it is determined that the existing template is not representative enough. At this time, all feature vectors in the cluster are weighted and averaged with the feature vectors of the original template. The weight coefficients of each feature dimension in the template are adjusted or the center vector of the template is updated so that the template can more accurately cover the actual distribution range of the special working condition.

[0092] The specific method for triggering alarm signals and locking fault characteristic data is as follows: the fault type corresponding to the final diagnosis result, the event fragment data at the time of the fault occurrence, and the test results that led to the fault determination are associated and stored in the fault database. At the same time, the alarm information is sent to the back-end terminal through the communication module, and the fault characteristic data is recorded for rapid matching of similar faults in the future.

[0093] It should be further explained that, in the specific implementation process, when the health assessment model determines that the equipment is in a fault state after comprehensively analyzing the first test result, the second test result, and its own wear trend, the system immediately triggers the alarm signal generation program. This program first determines the fault type based on the specific test result that led to the fault determination. The fault type includes at least one or a combination of electrical parameter out-of-tolerance faults, vibration characteristic abnormal faults, positioning accuracy out-of-tolerance faults, lubrication failure faults, and mechanical wear step faults.

[0094] The system then extracts all raw data within a preset time window before and after the fault occurrence from the data packet of the current event segment. This raw data is packaged with the event segment's condition label, timestamp, comparison data of the actual values ​​of each feature parameter in the first test result with the upper and lower limits of the dynamic envelope, comparison data of the specific value of the reconstruction error in the second test result with the preset threshold, the historical slope value of the wear trend curve in the health assessment model, and the current deviation value to generate a structured fault feature data record. This record is stored in the fault database of the local storage or cloud server. When storing, the system automatically assigns a unique fault number to the record and establishes a multi-level index based on the fault type.

[0095] Meanwhile, the central control system sends alarm information to the back-end terminal through the communication module. The alarm information includes at least the fault number, fault type, time of fault occurrence, and suggested handling measures. After receiving the alarm information, the back-end terminal highlights it on the monitoring interface and pops up a detailed fault report entry.

[0096] During the subsequent operation of the equipment, when a new event segment arrives and the first or second inspection result is abnormal, the system will prioritize the rapid matching of the feature data of the event segment with the existing fault feature records in the fault database. The matching method adopts one of the feature vector Euclidean distance or cosine similarity. If the matching degree exceeds the preset threshold, the corresponding fault type will be directly output as a warning information, and the processing suggestions in the historical fault records will be displayed in conjunction with it to realize the rapid identification of similar faults and the reuse of experience.

[0097] The method is also applicable to fault detection of other mechanical equipment with flipping mechanisms, including vehicle loading platforms in automated parking garages, flip-up boarding bridges, and cargo loading and unloading platforms. The application method is to map the operating characteristic parameters of the corresponding equipment to the collection dimensions of multi-source heterogeneous data, and to adapt the model according to the standard working condition template of the equipment.

[0098] It should be further explained that, in the specific implementation process, this method is not only applicable to the fault detection of the flip-type parking equipment itself, but can also be extended to other mechanical equipment with flipping mechanisms through parameter mapping and model adaptation, specifically including vehicle loading platforms of automated parking garages, flip-type boarding bridges, and cargo loading and unloading platforms.

[0099] When applied to the vehicle platform of a multi-level parking garage, the dimensions of the multi-source heterogeneous data collection are redefined based on the structural characteristics and operating logic of the vehicle platform. Among them, the electrical parameter data corresponds to the three-phase current and voltage values ​​of the vehicle platform lifting motor, the mechanical vibration data corresponds to the vibration spectrum characteristic values ​​at the connection between the vehicle platform guide rail and the transmission chain, the position angle data corresponds to the real-time angle change curve of the vehicle platform from the horizontal standby position to the vehicle storage and retrieval position, and the environmental compensation data includes the ambient temperature value and the lubrication status parameters obtained by monitoring the thickness of the guide rail grease film through ultrasonic sensors.

[0100] When applied to flip-type boarding bridges, the dimensions of multi-source heterogeneous data acquisition are redefined based on the rotation and telescopic mechanisms of the boarding bridge. Among them, electrical parameter data corresponds to the current and voltage values ​​of the boarding bridge's rotating and telescopic motors; mechanical vibration data corresponds to the vibration spectrum characteristic values ​​at the connection between the boarding port and the bridge body; position and angle data corresponds to the angle change curve of the boarding port's rotation in the horizontal plane and the length change curve of the bridge body's telescopic extension; and environmental compensation data includes ambient temperature and wind speed values, as well as lubrication status parameters monitored by ultrasonic sensors on the lubrication status of the rotating support bearings.

[0101] When applied to freight loading and unloading platforms, the dimensions of multi-source heterogeneous data collection are redefined based on the platform's lifting and tilting mechanisms. Among them, electrical parameter data corresponds to the current and voltage values ​​of the platform's lifting oil pump motor and tilting plate drive motor; mechanical vibration data corresponds to the vibration spectrum characteristic values ​​at the connection between the platform hinge and the hydraulic cylinder; position and angle data corresponds to the real-time angle change curves of the platform platform rising and falling from a horizontal position to a specified height and the tilting plate flipping from a retracted position to an unfolded position; and environmental compensation data includes ambient temperature values ​​and lubrication status parameters monitored by ultrasonic sensors to assess the hydraulic oil quality and hinge lubrication status.

[0102] In the above application scenarios, the central control system first initializes the basic parameters of the first and second inspection units based on the standard operating condition template library of the corresponding equipment. This standard operating condition template library is pre-built based on the equipment's design parameters and factory test data, and includes general operating condition templates as well as unique operating condition templates generated by subsequent self-learning.

[0103] Subsequently, the control system continuously collects multi-source heterogeneous data from the corresponding equipment and executes the same event segment cutting, parallel verification, arbitration verification, health verification, and dynamic update process. Among them, the event segment cutting redefines the start and end judgment rules of the start-up impact segment, uniform speed operation segment, arrival deceleration segment, and stationary standby segment according to the operating stage of the corresponding equipment. The arbitration logic of the arbitration verification unit fine-tunes the threshold based on the historical fault data and operating experience of the corresponding equipment. The Weibull distribution parameters in the health assessment model are recalibrated according to the design life and wear mechanism of the mechanical components of the corresponding equipment. Finally, the method achieves seamless migration and adaptive operation between different types of flipping mechanical equipment.

[0104] The multi-sensor threshold self-learning method of the flip-type berth equipment achieves refined identification of equipment operating status and dynamic threshold generation by constructing a multi-level parallel analysis architecture consisting of a first-level verification unit, a second-level verification unit, and a health assessment model, and by introducing an arbitration verification unit to perform no less than two cross-validations and iterative optimizations.

[0105] This method can effectively distinguish between parameter drift caused by changes in ambient temperature, fluctuations in lubrication conditions, or natural aging of machinery and actual fault characteristics. It avoids frequent false alarms triggered by fixed thresholds due to seasonal changes or equipment wear, while ensuring that sudden mechanical damage or electrical abnormalities are captured in a timely manner, thereby improving the accuracy of fault detection and the continuity of equipment operation.

[0106] This method feeds back the special working condition data marked by the arbitration verification unit to the standard working condition template library for clustering and updating, and uses the normal aging data determined by the health assessment model to dynamically adjust the statistical baseline of the statistical process control model, so that the threshold generation mechanism has continuous self-learning and self-adaptation capabilities, reducing the workload of manual intervention in adjusting the threshold.

[0107] Furthermore, this method enables rapid identification and experience reuse of similar faults through the associated storage and rapid matching of fault feature data. It can also be extended to various tipping mechanical equipment such as vehicle loading platforms in automated parking garages, flip-up boarding bridges, and cargo loading and unloading platforms through parameter mapping and model adaptation, thereby improving the versatility and engineering application value of the detection method.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-sensor threshold self-learning method for a flip-type berth device, characterized in that, Includes the following steps: S1: Collect multi-source heterogeneous data of the flip-type berth equipment at multiple preset monitoring points. The multi-source heterogeneous data includes at least electrical parameter data, mechanical vibration data, position angle data, and environmental compensation data. S2: Perform time-series segmentation on the collected multi-source heterogeneous data, dividing it into several event segments with physical meaning according to the equipment operation stage; S3: Input each event fragment simultaneously into the first and second verification units for parallel analysis; The first testing unit performs dynamic envelope analysis on the event segment based on a statistical process control model to generate the first testing result. The second verification unit performs similarity matching between the event fragment and the preset standard working condition template based on the working condition matching model, and generates a second verification result; S4: Input the first and second test results into the arbitration verification unit. The arbitration verification unit performs no less than two cross-validations and iterative optimizations on the two test results according to the preset arbitration logic to generate an intermediate judgment result. S5: Input the intermediate judgment results into the health assessment model for third-level verification. The health assessment model generates the final diagnosis result based on mechanical wear trend analysis. S6: If the final diagnosis result is a deviation from normal operating conditions, the statistical baseline parameters of the first inspection unit and / or the standard operating condition template of the second inspection unit will be dynamically updated based on the result. If the final diagnosis indicates a fault, an alarm signal will be triggered and the fault characteristic data will be locked.

2. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The electrical parameter data in the multi-source heterogeneous data includes the three-phase current and voltage values ​​of the drive motor; the mechanical vibration data includes the vibration spectrum characteristic values ​​of key bearing parts; the position angle data includes the real-time change curve of the flip plate rotation angle; and the environmental compensation data includes at least the ambient temperature value and lubrication state parameters, which are obtained by monitoring the oil film thickness of key transmission components using ultrasonic sensors.

3. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 2, characterized in that: The specific method for dividing the equipment into several physically meaningful event segments based on the equipment's operating phase is as follows: using sliding time window technology, based on the start / stop signal of the drive motor and the rate of change of the flip plate's angular velocity, the continuous data stream is sequentially divided into start-up impact segment, uniform speed flipping segment, arrival deceleration segment, and stationary standby segment, and timestamps and operating condition tags are added to each segment.

4. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The specific process of the first verification unit performing dynamic envelope analysis is as follows: statistical analysis is performed on event segments under the same working condition label whose historical running number reaches a preset number, the mean and standard deviation of each characteristic parameter are calculated, and the mean plus or minus n times the standard deviation is used as the upper and lower limits of the dynamic envelope, and the value of n is adaptively adjusted according to the number of times the equipment runs.

5. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The working condition matching model in the second verification unit is an autoencoder structure built on a long short-term memory network. The specific way it performs similarity matching is as follows: input the current event segment into the autoencoder to calculate the reconstruction error value. If the reconstruction error value is less than the preset threshold, it is determined to match the standard working condition template; otherwise, it is determined to be template mismatch. The standard operating condition template includes a preset general operating condition template and a self-learning generated special operating condition template. The special operating condition template is generated based on the clustering of event fragments marked as special operating conditions in historical operating data.

6. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The cross-validation and iterative optimization process performed by the arbitration verification unit includes: First round of arbitration: If the first inspection result is determined to be out of tolerance while the second inspection result is determined to be in compliance with the working condition template, the arbitration unit will not generate an alarm command, but will mark the current event segment as special working condition data and output it to the standard working condition template library for template update; If the first test result is normal but the second test result is template mismatch, the arbitration unit triggers a mechanical characteristic drift check instruction, packages the current event fragment and its test result, and sends it to the health assessment model. If both the first and second inspection results are deemed abnormal, the arbitration unit will directly generate a suspected fault signal and proceed to the second round of verification.

7. The multi-sensor threshold self-learning method for a flap-type berth device according to claim 6, characterized in that: The health assessment model is based on the Weibull distribution. The specific method for its third verification is as follows: receiving event fragment data sent by the arbitration unit, extracting feature parameters related to mechanical wear, and calculating the deviation between the current wear level value and the historical wear trend curve. If the deviation is within the preset gradual range, it is determined to be normal aging and the event segment is fed back as new baseline data to the first test unit for statistical parameter update; If the deviation exceeds the gradual range and exhibits a step-like characteristic, then the result of the first test and the result of the second test are combined to determine that it is a fault state.

8. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The specific method for dynamically updating the statistical baseline parameters of the first-level test unit is as follows: the feature values ​​of new event segments that the health assessment model determines to be normal aging are included in the historical dataset of the corresponding working condition label, and the mean and standard deviation are recalculated so that the dynamic envelope adaptively drifts with the degree of equipment aging. The specific method for dynamically updating the standard working condition template of the second-level inspection unit is as follows: the event fragments marked as special working conditions by the arbitration unit are clustered to generate a new working condition template, or the feature weights of the original working condition template are adjusted.

9. The multi-sensor threshold self-learning method for a flip-type berth device according to claim 1, characterized in that: The specific method for triggering the alarm signal and locking the fault feature data is as follows: the fault type corresponding to the final diagnosis result, the event fragment data at the time of the fault occurrence, and the test result that led to the fault determination are associated and stored in the fault database. At the same time, the alarm information is sent to the back-end terminal through the communication module, and the fault feature data is recorded for rapid matching of similar faults in the future.

10. The multi-sensor threshold self-learning method for a flap-type berth device according to claim 1, characterized in that: The method is also applicable to fault detection of other mechanical equipment with flipping mechanisms, including vehicle loading platforms in automated parking garages, flip-up boarding bridges, and cargo loading and unloading platforms. The application method is to map the operating characteristic parameters of the corresponding equipment to the acquisition dimension of the multi-source heterogeneous data, and perform model adaptation based on the standard working condition template of the equipment.

Images