Multi-sensor synchronous calibration jig and method

Abstract

The invention relates to a multi-sensor synchronous calibration jig and method. The precision, efficiency and stability of multi-sensor calibration can be improved. During specific operation, firstly, the multi-sensor array is accurately installed on the jig intelligent self-adaptive clamp and connected with an interface, and a system is electrified for self-inspection and uploads data information; then, starting a multi-dimensional state sensing network to collect environmental parameters, detecting the in-place state of the sensor and obtaining a self-inspection signal, and performing initial state evaluation; predicting the state of the actuating mechanism before action in state calibration and iteratively adjusting action parameters; finally, an effective parameter data point set and metadata thereof are input, traditional calibration parameter calculation is carried out, and state metadata serves as a covariable to be input into the error compensation model; according to the method, the accuracy, stability and reliability of sensor calibration are effectively improved through comprehensive evaluation of the initial state, dynamic adjustment of parameters, timely processing of abnormal data and the like, the calibration error is reduced, and the subsequent performance of the sensor is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sensor calibration, in particular to a multi-sensor synchronous calibration jig and method. BACKGROUND

[0002] In the field of modern industrial production and scientific research, multi-sensor systems have become a key technology for achieving high-precision measurement, monitoring and control. For example, in intelligent manufacturing, multi-sensors are used to accurately perceive the state of jigs and workpieces, helping to automate production processes; in the field of aerospace, multi-sensors work together to obtain all-around information of aircraft. To ensure the accuracy and reliability of multi-sensor system data, precise calibration is essential. Traditional multi-sensor calibration jigs and methods can basically meet the needs in relatively simple production environments, usually using manual operation and fixed processes to complete calibration.

[0003] However, as production scenarios become more complex, existing calibration jigs and methods have highlighted problems. The precision is low when installing sensors, and the interface connection is unstable, affecting the starting accuracy of calibration; the system self-checking function is limited, making it difficult to comprehensively detect the state of the sensor. State perception and parameter adjustment lack consideration of environmental dynamics and real-time state of the sensor, and the action parameters of the execution mechanism cannot be optimized in real time, resulting in large calibration errors. The traditional calibration parameter calculation method is fixed and does not fully utilize state metadata to compensate for errors. Data collection lacks synchronization and real-time analysis, and the ability to handle abnormal data is weak, affecting the calibration results. SUMMARY

[0004] The present application provides a multi-sensor synchronous calibration jig and method to improve the precision, efficiency and stability of multi-sensor calibration, and to adapt to the needs of complex and changing industrial production and scientific research scenarios.

[0005] The technical solution of the present application to solve the above technical problems is as follows: a multi-sensor synchronous calibration method, comprising: S1, accurately install a multi-sensor array to a jig intelligent adaptive clamp and connect various interfaces, power on the system for self-checking, and upload multi-sensor data information in the system; S2, start a multi-dimensional state perception network to collect environmental parameters, detect the in-place state of the sensor and obtain self-checking signals, and perform initial state evaluation on the collected data; S3, in state calibration, predict state iteration before the action of the jig execution mechanism to adjust the action parameters of the jig execution mechanism; S4, input the data point set and its metadata of the effective parameters, perform traditional calibration parameter calculation, and input the state metadata as a covariant into an error compensation model.

[0006] Preferably, the S3 comprises: 3A, acquire current environment and sensor state, predict the impact of performing actions, dynamically generate or adjust action parameters; 3B, the jig control execution mechanism accurately acts according to the fine tuning parameters, and continuously monitors the relevant states of the execution process; 3C, confirm that the execution is in place and the stimulation is stable, collect sensor real-time microstate indicators and environment indicators, and judge whether each indicator meets the standard in real time; 3D, after receiving the confirmation trigger instruction, the main controller synchronously sends a trigger pulse, and the sensor synchronously collects data and attaches state and environment snapshot metadata; 3E, quickly analyze the collected data online, and if it meets the standard, mark it as "collected", otherwise, mark it as failed and immediately retry collection according to the strategy.

[0007] Preferably, the real-time judgment of whether each indicator meets the standard comprises: According to the position sensor feedback information of the execution mechanism, it is confirmed whether the execution mechanism reaches the target position; the change of the sensor measurement value is observed to determine whether the stimulation is stable; if the execution is in place and the stimulation is stable, the initialization collection module is started to collect the real-time microstate indicators and environment indicators of the sensor; the collected data is analyzed in real time, the statistics of each indicator are calculated, and compared with the preset calibration standard; if each indicator meets the standard, a trigger instruction is sent to notify the main controller to prepare for data collection. If it does not meet the standard, retry, fine tuning compensation or skip are taken according to the specific situation.

[0008] Preferably, after S4, it further comprises: S41, identify key degradation factors affecting the stability of the sensor and the calibration jig, and establish a preliminary degradation model framework; S42, feature extraction is performed on real-time monitoring data, degradation features are associated with calibration results to generate data pairs, and an incremental learning algorithm is used to update degradation model parameters online; S43, set a prediction time window, predict the future state of the key degradation factor based on the updated model, and map its influence on the calibration performance; S44, calculate the risk probability of the key degradation factor exceeding the acceptable range; S45, define a maintenance action library according to the risk probability and predict the maintenance effect.

[0009] Preferably, S44, calculating the risk probability of the key degradation factor exceeding the acceptable range, comprises: Set the upper limit U and the lower limit L of the key degradation factor, the acceptable range is [L, U], and exceeding this range is considered as a risk; use the updated degradation model to predict the distribution of the key degradation factor within the future time window; set the predicted value to follow a normal distribution, with mean μ and standard deviation σ, and calculate the probability of exceeding the upper limit. is the cumulative distribution function of the standard normal distribution, Y is a random variable under the normal distribution; the probability of exceeding the lower limit is calculated: , then the total risk probability is: The prediction time window is divided into multiple time steps, the risk probability is calculated for each time step, and the cumulative effect within the time window is considered.

[0010] Preferably, the prediction key degradation factor comprises: Collecting current and historical data before the current time; pre-processing the data to ensure the quality of the input data; using the updated degradation model to predict the state of the key degradation factor in the future time window; for each time step, calculating the predicted value of the key degradation factor and its confidence interval; based on the model and historical operation data, defining the mapping relationship between the key degradation factor and the calibrated performance; inputting the predicted key degradation factor value into the mapping function to calculate the corresponding calibrated performance prediction value, considering the confidence interval of the key degradation factor, and calculating the prediction interval of the calibrated performance.

[0011] Preferably, the key degradation factor further comprises: The formula of the key degradation factor prediction model is: , wherein, is the predicted value of the key degradation factor at time , and are model parameters that have been updated by incremental learning and obtained by least squares method; for each time step in the future time window, ; the calibrated performance mapping function is , wherein P(y) is the calibrated performance, y is the value of the key degradation factor, and c and d are mapping parameters; the predicted key degradation factor value is substituted into the mapping function to calculate the predicted value of the calibrated performance .

[0012] Preferably, the S44 comprises: S441, each client trains a local risk probability model using its own private data, and based on the trained local model, simulates a series of predicted risk probability results to form a knowledge package; wherein the client refers to each group of multi-sensor synchronous calibration fixture combination; S442, the client encrypts the generated knowledge package, and the system center receives the encrypted knowledge package uploaded by each client and performs homomorphic encryption processing on the model output part in the knowledge package; S443, the system center securely aggregates the encrypted prediction results from different clients for the same or similar input features; ​S444, the system center trains a global risk probability model by using the encrypted result after security aggregation and combining shared operation data; S445, the system center applies the trained global model to the same input feature set to generate a global predicted risk probability result, and forms a global knowledge package with the corresponding input features; S446, when the client has a similar operation request, the system center distributes the encrypted global knowledge package to the client; S447, after receiving the encrypted global knowledge package, the client decrypts and processes it, and the client updates the local risk probability model by combining local data and the predicted result in the global knowledge package.

[0013] Preferably, the S441 comprises: Each client collects its own private data, extracts features related to risk prediction, and divides the data into a training set and a validation set; the client uses a machine learning model to train the selected model using the training set data, the model learns the relationship between the features of the local data and the risk probability, and obtains a local risk probability model Wherein is the input feature vector of the client i, is the predicted risk probability output by the model; the client simulates a series of input features similar to the actual operation scenario based on historical operation data, inputs the simulated input features into the trained local model , calculates the corresponding predicted risk probability, and forms a knowledge package with the simulated input features and the corresponding predicted risk probability.

[0014] The application also provides a multi-sensor synchronous calibration jig, which comprises An initialization acquisition module is used for accurately installing a multi-sensor array to a jig intelligent adaptive clamp and connecting various interfaces, performing self-checking after power-on, and uploading multi-sensor data information in the system; A perception evaluation module is used for starting a multi-dimensional state perception network to collect environmental parameters, detecting the in-place state of the sensor and obtaining a self-checking signal, and performing initial state evaluation on the collected data; A calibration adjustment module is used for predicting the state iteration of the jig execution mechanism before the action of the jig execution mechanism in state calibration, and adjusting the action parameter strategy of the jig execution mechanism; A calibration compensation module is used for inputting a set of data points and metadata of effective parameters, performing traditional calibration parameter calculation, and inputting the state metadata as a covariant into an error compensation model.

[0015] The application has the following beneficial effects: By precisely installing a multi-sensor array and uploading key data information, a multi-dimensional state perception network is used to comprehensively evaluate the initial state of the system, and according to the evaluation results, flexible processing is carried out, the action parameters of the actuator are dynamically adjusted in the state calibration, and the data is collected after the execution is in place and the stimulation is stable. And quickly analyze and process the collected data, and finally use the effective data and metadata for calibration and error compensation. This series of operations effectively improves the accuracy, stability and reliability of sensor calibration, can timely discover and handle abnormal situations, reduce calibration errors, and ensure the performance of sensors in subsequent applications.

[0016] By real-time monitoring and intelligent analysis of key degradation factors of multi-sensor and calibration fixture, high-precision synchronous calibration and performance prediction are realized. This method significantly improves the accuracy and efficiency of calibration, effectively reduces the calibration error caused by sensor drift or fixture degradation, and ensures the stability and reliability of long-term operation of the equipment. At the same time, through the incremental learning algorithm to dynamically update the calibration model, it can quickly adapt to environmental changes and equipment aging, reduce maintenance costs and downtime, and provide an efficient and reliable solution for industrial automation and precision measurement fields.

[0017] Through the collaborative work of the client and the system center, data sharing and knowledge fusion across devices are realized. Each client trains a local model using private data and generates a knowledge package, the system center processes and aggregates the knowledge package, trains a global model based on shared data, and feeds back global knowledge to the client to update the local model. This collaborative mode effectively integrates the data and knowledge of multiple devices, improves the accuracy and comprehensiveness of risk prediction, enhances the adaptability of devices in complex environments, and promotes the complementary and collaborative development across devices. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a multi-sensor synchronous calibration method of the present application; Figure 2 A structural block diagram of a multi-sensor synchronous calibration fixture of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] In the description of the present application, the terms "first", "second", are used only for descriptive purposes and are not to be construed as indicating or implying relative importance or an indicated number of technical features. Thus, features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise expressly and specifically limited.

[0021] In the description of the present application, the term "for example" is used to indicate "as an example, illustration, or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given so that any person skilled in the art can implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can realize the present application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope consistent with the principles and characteristics disclosed in the present application.

[0022] Embodiment 1: Figure 1 is a flowchart of a multi-sensor synchronous calibration method according to an embodiment of the present application, comprising the following steps: S1, accurately install the multi-sensor array to the intelligent adaptive fixture of the fixture and connect various interfaces, power on the system for self-checking, upload multi-sensor data information in the system.

[0023] Among them, the multi-sensor data information includes sensor type, calibration accuracy, health state baseline model and calibration specification requirements, and the sensors in the present application also refer to the plurality of sensors installed on the fixture.

[0024] Specifically, the multi-sensor array is placed on the mounting position of the intelligent adaptive fixture according to the predetermined layout and direction. Pay attention to the pins, interfaces and other parts of the sensor not to be squeezed or collided. According to the design of the sensor and the fixture, use the appropriate fixing method to firmly install the sensor on the fixture. During installation, ensure the positional accuracy of the sensor to avoid deviation or tilt, and require the positional deviation to be strictly controlled within the allowable tolerance range. According to the corresponding relationship of the interface, connect the power interface of the sensor with the power supply of the system, ensure the correct polarity of the power supply, and avoid reverse connection to cause damage to the sensor. At the same time, check whether the power voltage meets the rated working voltage requirement of the sensor. Connect the data transmission interface of the sensor with the initialization acquisition module or controller of the system. For serial communication, ensure that the baud rate, data bits, stop bits, and check bits are correctly set; for USB interface or Ethernet interface, ensure that the driver is installed correctly and the network connection is normal. If the sensor has a signal input / output interface, connect it with the corresponding signal source or other equipment according to actual needs. During connection, pay attention to signal matching and isolation to avoid signal interference or damage to equipment. After confirming that all interfaces are correctly connected, power on the system step by step according to the system's operation procedures. Through the data interface of the system, upload the prepared multi-sensor data information to the system.

[0025] S2, start the multi-dimensional state perception network to collect environmental parameters, detect the in-place state of the sensor, and obtain self-check signals, and perform initial state evaluation on the collected data.

[0026] Among them, the super serious threshold value stops the calibration alarm, the slight deviation is pre-compensated, and the state is good to start the calibration.

[0027] Specifically, the environmental sensors in the multi-dimensional state perception network start collecting environmental parameters according to the preset sampling frequency. The initialization acquisition module converts the analog signals collected by the sensor into digital signals and performs preliminary filtering and preprocessing to remove noise and interference. The system detects whether each sensor is in place through a hardware interface or a communication protocol. For example, for sensors using digital interfaces, a specific query instruction can be sent to determine whether the sensor is in place according to the response of the sensor; for sensors with analog interfaces, the level signal or impedance change of the interface can be detected to determine. After starting or receiving the self-check instruction, the sensor will perform internal self-checking and send the self-checking result in the form of a signal to the system. The self-checking signal may include information such as the working mode, measurement range, accuracy, and fault code of the sensor.

[0028] According to the results of data analysis, the initial state of the system is divided into three cases: super serious, slight deviation, and good state.

[0029] If the initial state evaluation result of the system is super serious, that is, there is a serious fault or abnormal situation, the system will immediately suspend the calibration process and issue an alarm signal. At the same time, the system will record the time, place, fault type and other information of the fault. If the initial state evaluation result of the system is slight deviation, that is, there are some small errors or unstable factors, but it does not affect the normal calibration, the system will automatically perform pre-compensation. The pre-compensation method can be selected according to the specific situation (for example, for small range deviation of temperature sensor, the compensation can be performed by adjusting the correction coefficient of the measured value; for small fluctuations of the signal, the smoothing processing can be performed by filtering algorithm). If the initial state evaluation result of the system is good, that is, all sensors work normally and the environmental parameters are within a reasonable range, the system will start the calibration process and enter the subsequent calibration steps.

[0030] S3, in the state calibration, the state iteration before the action of the jig execution mechanism is predicted, and the action parameter adjustment strategy of the jig execution mechanism is performed.

[0031] Specifically, 3A, the current environment and sensor state are obtained, the influence of the execution action is predicted, and the action parameters are dynamically generated or adjusted.

[0032] Before step 3A is performed, a multi-dimensional state perception network is used to collect a large amount of historical data about sensor state, environmental parameters and jig execution mechanism action. The collected data is labeled to clearly indicate the corresponding sensor target state and the action to be taken by the execution mechanism. For example, in a jig system for calibrating an acceleration sensor, the acceleration values measured by the sensor under different rotation angles and speeds of the execution mechanism in different temperature and humidity environments need to be recorded. When the sensor is expected to measure the acceleration in a specific direction, the angle and speed at which the execution mechanism needs to be rotated are labeled. A neural network is used to automatically learn the complex mapping relationship between the sensor state and the execution mechanism action, and a state-action model is constructed.

[0033] Various sensors in the multi-dimensional state perception network collect real-time environmental parameters (temperature, humidity, air pressure) and sensor state information. The initialization acquisition module converts the analog signals collected into digital signals and performs preliminary filtering and preprocessing to remove noise and interference.

[0034] The collected environmental parameters and sensor state information are integrated to form a complete state data set.

[0035] Using the pre-established state-action model, the current state data set is input to predict the influence of different actions of the jig execution mechanism on the sensor state and the environment. For example, if the execution mechanism is a mechanical arm used to adjust the position of the sensor, the influence of different moving distances and directions on the sensor measurement value is predicted.

[0036] According to the prediction results and the calibration target (reach a specific measurement value or working state), the action parameters of the jig actuator are dynamically generated or adjusted, such as the speed, amplitude, direction and duration of the action. Optimal action parameters are determined using optimization algorithms to minimize the error between the prediction results and the calibration target.

[0037] 3B, the jig control actuator accurately acts according to the fine tuning parameters, and the execution process related state is continuously monitored.

[0038] The jig control system generates corresponding action instructions according to the action parameters generated or adjusted in step 3A, and sends the instructions to the actuator through the communication interface.

[0039] After receiving the action instructions, the actuator accurately performs the action according to the instructions.

[0040] During the execution of the actuator, the multi-dimensional state perception network continuously monitors the state of the actuator and the state of the sensor.

[0041] If the actuator action is found to be abnormal or the sensor state is abnormal, an alarm signal is sent in time.

[0042] 3C, confirm that the execution is in place and the stimulation is stable, collect sensor real-time microstate indicators and environmental indicators at high speed, and judge whether each indicator meets the standard in real time.

[0043] Among them, if the indicator meets the standard, a trigger instruction is sent, and if it does not meet the standard, retry / fine tuning compensation / skip.

[0044] According to the feedback information of the position sensor of the actuator, it is confirmed whether the actuator reaches the target position. At the same time, observe the change of the sensor measurement value, judge whether the stimulation is stable (the measurement value fluctuation range in a certain time is less than the set threshold).

[0045] If the execution is in place and the stimulation is stable, start the initialization collection module to collect sensor real-time microstate indicators and environmental indicators at a high sampling frequency.

[0046] Real-time analysis is performed on the collected data, and the average value, standard deviation, maximum value, minimum value and other statistical quantities of each indicator are calculated and compared with the preset calibration standard.

[0047] If all indicators meet the standard, a trigger instruction is sent to notify the main controller to prepare for data collection. If it does not meet the standard, according to the specific situation, take measures such as retry (re-execute steps 3A-3C), fine tuning compensation (adjust the action parameters of the actuator and try again), or skip (if the indicator does not meet the standard, it has little effect on subsequent calibration, and the current step can be skipped).

[0048] It should be noted that after reaching the standard, the trigger parameters such as trigger time, trigger condition, etc. can be fine-tuned according to actual situation.

[0049] 3D, after receiving the trigger instruction, the main controller synchronously sends out a trigger pulse, and the sensor synchronously collects data and attaches state and environmental snapshot metadata.

[0050] The main controller receives the trigger instruction sent in step 3C.

[0051] The main controller synchronously sends out a trigger pulse to inform the sensor to start collecting data.

[0052] After receiving the trigger pulse, the sensor immediately starts collecting data and simultaneously records the current state and environmental snapshot metadata, such as the working mode of the sensor, measurement values, environmental temperature, humidity, etc.

[0053] The collected data and metadata are stored and transmitted in a predetermined format.

[0054] 3E, online analysis of collected data, if passed, mark as "collected", if abnormal, mark as failed and immediately retry collection according to the strategy.

[0055] After the system receives the data and metadata collected by the sensor, it performs preprocessing such as noise removal and data format conversion.

[0056] The collected data is quickly analyzed online using a preset threshold judgment algorithm to determine whether the data meets the calibration requirements.

[0057] If the data passes the analysis, it is marked as "collected" and stored in the database. If the data is abnormal, it is marked as failed and immediately reattempted according to the preset strategy (reexecute steps 3A-3E, adjust collection parameters and try again).

[0058] Among them, the qualified threshold of a certain data index is [lower limit value, upper limit value], and the collected data value is d, then the formula to judge whether the data is qualified is: lower limit value ≤ d ≤ upper limit value, if the data does not meet this condition, it is judged as abnormal.

[0059] S4, input the data point set of valid parameters and its metadata, perform traditional calibration parameter calculation, and input the state metadata as a covariant into the error compensation model.

[0060] Specifically, the data point set of valid parameters is selected from the previously collected data. Valid parameters usually refer to data that meets certain quality standards. For example, in sensor calibration, invalid data points caused by excessive noise interference or abnormal sensor operation are removed.

[0061] Metadata related to the data points of the effective parameters is collected, including acquisition time, ambient temperature, humidity, and working mode information of the sensor.

[0062] According to the type and characteristics of the sensor, a linear model calibration model (y=ax+b, y is the output value of the sensor, x is the physical quantity to be measured, and a and b are calibration parameters) is selected, and the least square method is used to calculate the parameters of the calibration model.

[0063] The original data is calibrated using the calculated calibration parameters to obtain a calibrated value (y' = a x + b) The calibration error (e = y - y') is calculated, which is the difference between the original measurement value and the calibrated value.

[0064] The state metadata is input as a covariate into a machine learning error compensation model, for example, if the state metadata includes ambient temperature T and humidity H, the input of the error compensation model can be represented as (y = f(T, H, e)) The output is the error (e) The error compensation model is trained using a data set containing the input covariates and the error.

[0065] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: By precisely installing a multi-sensor array and uploading key data information, the initial state of the system is comprehensively evaluated using a multi-dimensional state perception network, and the evaluation results are flexibly processed. In the state calibration, the action parameters of the execution mechanism are dynamically adjusted to ensure that the data is collected after the execution is in place and the stimulation is stable, and the collected data is quickly analyzed and processed. Finally, the effective data and metadata are used for calibration and error compensation. This series of operations effectively improves the accuracy, stability and reliability of sensor calibration, can timely discover and handle abnormal situations, reduces calibration errors, and ensures the performance of the sensor in subsequent applications.

[0066] Embodiment 2 In embodiment 1, for the multi-sensor synchronous calibration process, a unified calibration model and process are used to handle the degradation of different sensors and fixtures. However, the key degradation factors of different sensors and calibration fixtures are significantly different, and the influence of these degradation factors on calibration performance varies. The unified model is difficult to accurately capture the complex relationship between each factor and calibration performance when dealing with diversified degradation characteristics, resulting in limited calibration accuracy and insufficient generalization ability. Due to the differences in the degradation patterns of different sensors and fixtures in history, the prediction and maintenance requirements for calibration performance are also different. Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0067] In some embodiments, after step S4, the method further comprises: ​S41, identify key degradation factors affecting the stability of the sensor and the calibration fixture, and establish a preliminary degradation model framework.

[0068] Among them, the key degradation factor refers to those factors that have a significant impact on the performance, reliability, accuracy or service life of the device during the operation of the fixture or system, and the change or degradation of which will cause the performance of the device to decline or fail. The key degradation factors of sensors can include zero drift, sensitivity decline, noise increase, temperature drift, etc. The key degradation factors of calibration fixtures can include mechanical wear, deformation, pollution, aging, etc.

[0069] Specifically, first, collect the historical operation data of the sensor and the calibration fixture, including but not limited to temperature, humidity, vibration, number of uses, maintenance records, calibration results, etc. Then, based on the analysis of the historical operation data, the factors that may affect the stability of the sensor and the calibration fixture are preliminarily screened out. Next, the relationship between each factor and the performance of the sensor and the calibration fixture is further analyzed by statistical analysis, and finally the key degradation factors are determined. According to the characteristics and data distribution of the key degradation factors, a degradation model is established. Based on the historical operation data, the initial values of the model parameters are set, and based on the type and parameter initialization of the degradation model, a preliminary degradation model framework is constructed, which can describe the change trend of the key degradation factors with time or number of uses, and the influence of such change on the performance of the sensor and the calibration fixture.

[0070] S42, feature extraction is performed on real-time monitoring data, degradation features are associated with calibration results to generate data pairs, and an incremental learning algorithm is used to update the degradation model parameters online.

[0071] Specifically, real-time operation data is collected through sensors and monitoring systems, features of the operation data are extracted, and the extracted degradation features are time-correlated with subsequent calibration results to form a "degradation feature-performance impact" data pair. For example, the vibration feature at a certain time is associated with the calibration error at that time. An incremental learning algorithm is used to update the model parameters in real time. For example, for a linear model, the weights are updated using online gradient descent. The prediction error (MSE, MAE) of the model is calculated, and the model confidence is calculated based on the error or other indicators (amount of data, feature significance).

[0072] S43, set a prediction time window, predict the future state of the key degradation factor based on the updated model, and map its influence on the calibration performance.

[0073] Specifically, according to the maintenance requirements and historical data period, the time range of prediction is set, and the time step of prediction is determined. First, the current time and the historical data before the current time are collected, including the monitoring values of the key degradation factors, environmental parameters, etc. Then, the data is preprocessed to ensure the quality of the input data. Next, the updated degradation model is used to predict the state of the key degradation factors in the future time window. For each time step, the predicted value of the key degradation factor and its confidence interval are calculated. Based on the model and historical operation data, the mapping relationship between the key degradation factor and the calibration performance is defined. The predicted value of the key degradation factor is input into the mapping function, and the corresponding predicted value of the calibration performance is calculated. Considering the confidence interval of the key degradation factor, the prediction interval of the calibration performance is calculated.

[0074] wherein the key degradation factor prediction model formula is: wherein, is the predicted value of the key degradation factor at time , and are model parameters that have been updated by incremental learning and obtained by least squares method; for each time step in the future time window, is calculated; the calibration performance mapping function is wherein P(y) is the calibration performance, y is the value of the key degradation factor, and c and d are mapping parameters. The predicted value of the key degradation factor is substituted into the mapping function to calculate the predicted value of the calibration performance .

[0075] S44, calculate the risk probability of the key degradation factor exceeding the acceptable range.

[0076] Specifically, the upper limit (U) and the lower limit (L) of the key degradation factor are set, and the acceptable range is [L, U], exceeding this range is considered as risk. Using the updated degradation model, the distribution of the key degradation factor in the future time window is predicted. Assuming that the predicted value follows a normal distribution, the mean is μ, and the standard deviation is σ, the probability of exceeding the upper limit is calculated: , is the cumulative distribution function of the standard normal distribution, Y is a random variable under this normal distribution, used to describe the possible value of the key degradation factor in the future time window, and the probability of exceeding the lower limit is calculated: , then the total risk probability is: The prediction time window is divided into multiple time steps, and the risk probability is calculated for each time step, and the cumulative effect in the time window is considered.

[0077] S45, define the maintenance action library according to the risk probability and predict the maintenance effect.

[0078] Specifically, first, a maintenance action library is defined, including preventive maintenance, corrective maintenance, and repair maintenance, and the risk probability range applicable to each maintenance action is defined. Then, the corresponding maintenance action is selected based on the risk probability. Next, the maintenance effect is predicted, and the key degradation factor distribution and risk probability after maintenance are calculated by establishing a mapping relationship between maintenance actions and key degradation factor improvement. Finally, the optimal maintenance action is selected according to the maintenance effect evaluation index (such as risk reduction amount, cost-benefit ratio, etc.).

[0079] The technical solutions in the embodiments of the application have at least the following technical effects or advantages: By real-time monitoring and intelligent analysis of the key degradation factors of multiple sensors and calibration fixtures, high-precision synchronous calibration and performance prediction are realized. This method significantly improves the accuracy and efficiency of calibration, effectively reduces the calibration error caused by sensor drift or fixture degradation, and ensures the stability and reliability of long-term operation of the equipment. At the same time, through the incremental learning algorithm, the calibration model is dynamically updated, which can quickly adapt to environmental changes and equipment aging, reducing maintenance costs and downtime, and providing an efficient and reliable solution for industrial automation and precision measurement fields.

[0080] Embodiment 3 In embodiment 2, each multi-sensor synchronous calibration fixture combination independently trains a local risk probability model using its own private data, and uploads the simulated generated knowledge package to the system center for global model training. However, the data distribution, operating environment, and equipment characteristics of different clients are quite different, resulting in uneven prediction ability of local models for risk probability. Moreover, the knowledge packages uploaded by different clients vary in data size, feature dimension, and prediction result accuracy. When the system center directly performs secure aggregation and global model training on these knowledge packages, it is difficult to fully consider these differences, making the global model have limited generalization ability when facing diversified actual operating scenarios, and unable to accurately adapt to the specific needs of different clients.

[0081] Therefore, the embodiments of the present application make certain optimizations on the basis of the above embodiments.

[0082] In some embodiments, after step S44, the method further includes: S441, each client trains a local risk probability model using its own private data, and based on the trained local model, simulates a series of predicted risk probability results to form a knowledge package.

[0083] Wherein, the client refers to each group of multi-sensor synchronous calibration fixture combinations.

[0084] Specifically, each client collects its own private data, which includes but is not limited to device operating parameters, environmental information, and historical risk event records. The collected data is cleaned to remove noise and outliers, feature engineering is performed to extract features related to risk prediction, such as feature extraction on time series data, and the data is divided into training and validation sets. The client uses a machine learning model to train the selected model using the training set data, adjusts the model parameters through optimization algorithms to optimize the performance of the model on the validation set. During the training process, the model learns the relationship between the characteristics of the local data and the risk probability, and obtains the local risk probability model wherein is the input feature vector of client i, is the predicted risk probability output by the model.

[0085] Based on historical operating data, the client simulates a series of input features similar to the actual operating scenario, inputs the simulated input features into the trained local model , and calculates the corresponding predicted risk probability. The simulated input features and the corresponding predicted risk probability form a knowledge package.

[0086] S442, the client encrypts the generated knowledge package, and the system center receives the encrypted knowledge packages uploaded by each client and performs homomorphic encryption processing on the model output part in the knowledge package.

[0087] wherein the model output part in the knowledge package is the predicted risk probability result.

[0088] S443, the system center securely aggregates the encrypted prediction results from different clients for the same or similar input features.

[0089] S444, the system center uses the securely aggregated encrypted results to train a global risk probability model in combination with shared operating data.

[0090] wherein the shared operating data is public industry data, historical experience data, etc.

[0091] S445, the system center applies the trained global model to the same input feature set to generate the result of the global predicted risk probability, and combines it with the corresponding input features to form a global knowledge package.

[0092] S446, when the client has a similar operating request, the system center distributes the encrypted global knowledge package to the client.

[0093] S447, after receiving the encrypted global knowledge package, the client decrypts it, and the client updates the local risk probability model in combination with the local data and the predicted results in the global knowledge package.

[0094] The technical solutions in the embodiments of the application have at least the following technical effects or advantages: Through the cooperation of the client and the system center, data sharing and knowledge fusion across devices are realized. Each client trains a local model using private data and generates a knowledge package, the system center performs security processing and aggregation on the knowledge package, trains a global model in combination with shared data, and feeds back global knowledge to the client to update the local model. This cooperative mode effectively integrates the data and knowledge of multiple devices, improves the accuracy and comprehensiveness of risk prediction, enhances the adaptability of devices in complex environments, and promotes the complementary advantages and cooperative development across devices.

[0095] Further, the embodiment of the application also provides a multi-sensor synchronous calibration jig.

[0096] Figure 2 is a structural schematic diagram of a multi-sensor synchronous calibration jig according to an embodiment of the application.

[0097] As shown in Figure 2 A multi-sensor synchronous calibration jig includes an initialization acquisition module, a perception evaluation module, a calibration adjustment module, and a calibration compensation module.

[0098] The initialization acquisition module is used to accurately install a multi-sensor array to a jig intelligent adaptive clamp and connect various interfaces, perform a self-check after power-on, and upload multi-sensor data information in the system; The perception evaluation module is used to start a multi-dimensional state perception network to collect environmental parameters, detect the in-place state of the sensor and obtain a self-check signal, and perform initial state evaluation on the collected data; The calibration adjustment module is used to predict the state iteration before the action of the jig execution mechanism in state calibration, and perform action parameter adjustment strategy of the jig execution mechanism; The calibration compensation module is used to input a data point set and metadata of effective parameters, perform traditional calibration parameter calculation, and input the state metadata as a covariant into an error compensation model.

[0099] It should be noted that other specific implementation contents of the multi-sensor synchronous calibration jig according to the embodiment of the application can refer to the multi-sensor synchronous calibration method described above.

[0100] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0101] Those skilled in the art will appreciate that embodiments of the present application can be devised for a variety of applications. It is intended that the present application be limited only by the scope of the appended claims, and it is intended that various modifications and alterations made by those skilled in the art be considered as within the scope of the present application. The embodiments of the present application will be described with reference to the attached drawings identified below.

[0102] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0103] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0105] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such modifications and variations as fall within the scope of the present application.

[0106] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A multi-sensor synchronous calibration method, characterized by, The method comprises: S1, accurately install a multi-sensor array to a jig intelligent adaptive clamp and connect various interfaces, system power self-check, upload multi-sensor data information in the system; S2, start a multi-dimensional state perception network to collect environmental parameters, detect the in-place state of the sensor and obtain a self-check signal, and perform initial state evaluation on the collected data; S3, in state calibration, predict state iteration before jig actuator action to adjust the action parameter strategy of the jig actuator; S4, input the data point set and its metadata of the effective parameters, perform traditional calibration parameter calculation, and input the state metadata as a covariant into an error compensation model.

2. The method of claim 1, wherein, The S3 comprises: 3A, obtain the current environment and sensor state, predict the execution action influence, and dynamically generate or adjust the action parameters; 3B, the jig control actuator accurately acts according to the fine tuning parameters, and continuously monitors the relevant states in the execution process; 3C, confirm that the execution is in place and the stimulation is stable, collect the real-time microstate indicators and environmental indicators of the sensor, and judge whether each indicator meets the standard in real time; 3D, after receiving the confirmation trigger instruction, the main controller synchronously sends a trigger pulse, the sensor synchronously collects data and attaches state and environmental snapshot metadata; 3E, quickly analyze the collected data online, and if the data meets the standard, mark it as "collected", otherwise, mark it as failed and immediately retry the collection according to the strategy.

3. The method of claim 2, wherein, The real-time judgment of whether each indicator meets the standard comprises: According to the position sensor feedback information of the actuator, confirm whether the actuator reaches the target position; observe the change of the sensor measurement value, judge whether the stimulation is stable; if the execution is in place and the stimulation is stable, start the initialization collection module to collect the real-time microstate indicators and environmental indicators of the sensor; perform real-time analysis on the collected data, calculate the statistics of each indicator, and compare them with the preset calibration standard; if each indicator meets the standard, send a trigger instruction to notify the main controller to prepare for data collection. If it does not meet the standard, retry, fine-tune compensation or skip according to the specific situation.

4. The method of claim 1, wherein, After S4, it further comprises: S41, identify the key degradation factors affecting the stability of the sensor and the calibration jig, and establish a preliminary degradation model framework; S42, extract features from real-time monitoring data, associate degradation features with calibration results to generate data pairs, and update degradation model parameters online using incremental learning algorithms; S43, set a prediction time window, predict the future state of the key degradation factors based on the updated model, and map their influence on the calibration performance; S44, calculate the risk probability of the key degradation factors exceeding the acceptable range; S45, define a maintenance action library according to the risk probability and predict the maintenance effect.

5. The method of claim 4, wherein, The S44, calculating the risk probability of the key degradation factors exceeding the acceptable range, comprises: Setting the upper limit U and the lower limit L of the key degradation factor, the acceptable range is [L, U], and exceeding this range is considered as risk; using the updated degradation model, predicting the distribution of the key degradation factor within the future time window; setting the predicted value to follow a normal distribution, with mean μ and standard deviation σ, calculating the probability of exceeding the upper limit: , is the cumulative distribution function of the standard normal distribution, and Y is the random variable under this normal distribution; calculating the probability of exceeding the lower limit: , then the total risk probability is: Divide the prediction time window into multiple time steps, calculate the risk probability for each time step, and consider the cumulative effect within the time window.

6. The method of claim 5, wherein, The prediction of the key degradation factors comprises: Collecting current time and historical data before; pre-processing the data to ensure the quality of the input data; using the updated degradation model to predict the state of the key degradation factor in the future time window; for each time step, calculate the predicted value of the key degradation factor and its confidence interval; based on the model and historical operation data, define the mapping relationship between the key degradation factor and the calibration performance; input the predicted key degradation factor value into the mapping function to calculate the corresponding calibration performance prediction value, consider the confidence interval of the key degradation factor, and calculate the prediction interval of the calibration performance.

7. The method of claim 6, wherein, The key degradation factor further comprises: The key degradation factor prediction model formula is: wherein, is the predicted value of the key degradation factor at time , and are model parameters that have been updated through incremental learning, and are obtained through least squares method; for each time step in the future time window, calculate ; the calibration performance mapping function is , wherein P(y) is the calibration performance, y is the value of the key degradation factor, and c and d are mapping parameters; the predicted value of the key degradation factor is substituted into the mapping function to calculate the predicted value of the calibration performance .

8. The method of claim 4, wherein, The S44 comprises: S441, each client trains a local risk probability model using its own private data, and based on the trained local model, simulates a series of predicted risk probability result groups to form a knowledge package; wherein the client refers to each group of multi-sensor synchronous calibration fixture combinations; S442, the client encrypts the generated knowledge package, and the system center receives the encrypted knowledge package uploaded by each client and performs homomorphic encryption processing on the model output part in the knowledge package; S443, the system center securely aggregates the encrypted prediction results from different clients for the same or similar input features; S444, the system center trains a global risk probability model using the securely aggregated encrypted results combined with shared operation data; S445, the system center applies the trained global model to the same input feature set to generate a global predicted risk probability result, and combines it with the corresponding input features to form a global knowledge package; S446, when the client has a similar operation request, the system center downloads the encrypted global knowledge package to the client; S447, after receiving the encrypted global knowledge package, the client decrypts it, and the client updates the local risk probability model based on the local data and the prediction result in the global knowledge package.

9. The method of claim 8, wherein, The S441 comprises: Each client collects its own private data, extracts features related to risk prediction, divides the data into training set and validation set; the client uses a machine learning model, uses the training set data to train the selected model, the model learns the relationship between the characteristics of the local data and the risk probability, and obtains the local risk probability model wherein is the input feature vector of the client i, is the predicted risk probability output by the model; based on the historical operation data, the client simulates a series of input features similar to the actual operation scene, inputs the simulation input features into the trained local model , calculates the corresponding predicted risk probability; the simulation input features and the corresponding predicted risk probability form a knowledge package.

10. A multi-sensor synchronous calibration jig applied to the multi-sensor synchronous calibration method according to any one of claims 1 to 9, characterized in that, The fixture comprises An initialization acquisition module is configured to accurately install a multi-sensor array to a fixture intelligent adaptive clamp and connect various interfaces, perform a self-check after the system is powered on, and upload multi-sensor data information in the system; A perception evaluation module is configured to start a multi-dimensional state perception network to collect environmental parameters, detect the in-place state of a sensor, and obtain a self-check signal, and perform initial state evaluation on the collected data; A calibration adjustment module is configured to predict state iteration before the action of a fixture execution mechanism in state calibration, and perform action parameter adjustment strategies for the fixture execution mechanism; A calibration compensation module is configured to input a data point set and metadata of effective parameters, perform traditional calibration parameter calculation, and input state metadata as a covariant into an error compensation model.

Images