Mathematical model-based multi-sensor dynamic calibration method and system

Abstract

The invention discloses a multi-sensor dynamic calibration method and system based on a mathematical model, and relates to the technical field related to computer systems, and the method comprises the steps: obtaining a historical weighing sample data set through building a dynamic weighing transmission structure; and training a dynamic disturbance filtering model according to the historical weighing sample data set and a weighing error label sample, wherein a weighing error label is an error between historical weighing transmission data and actual weight data of an article. The real-time weighing transmission data is acquired, the real-time weighing transmission data is sent to the dynamic disturbance filtering model, the weighing compensation data is acquired for compensation calibration, and the calibrated weighing transmission data is output. The technical problems of unstable weighing data and large error caused by factors such as object movement, transmission belt acceleration and vibration in the dynamic weighing process in the prior art are solved. The technical effects that a high-precision dynamic weighing result which is stable and close to the real weight is output is achieved, and the weighing accuracy and reliability of the system are improved are achieved.

Description

Technical Field

[0001] This application relates to the technical field of computer systems, and specifically to a multi-sensor dynamic calibration method and system based on a mathematical model. Background Art

[0002] In the fields of industrial production and logistics transportation, dynamic weighing, as an important process monitoring means, is widely used in scenarios such as item sorting, loading verification, and inventory management. Common dynamic weighing systems are mostly implemented by deploying weighing units on mobile platforms, which can complete weighing during the movement of items. However, due to factors such as the inertial effect of items in the mobile platform, platform vibration, and acceleration and deceleration changes, it will cause instability or even systematic deviation of the weighing readings, seriously affecting the accuracy and reliability of weighing data.

[0003] Therefore, in the prior art, there are technical problems of unstable weighing data and large errors during the dynamic weighing process due to factors such as item movement, conveyor belt acceleration, and vibration. Summary of the Invention

[0004] This application provides a multi-sensor dynamic calibration method and system based on a mathematical model, which solves the technical problems of unstable weighing data and large errors during the dynamic weighing process in the prior art due to factors such as item movement, conveyor belt acceleration, and vibration. It realizes high-precision dynamic weighing results that are stable and close to the actual weight through multi-sensor fusion and disturbance compensation based on Kalman filtering in a dynamic environment, and improves the weighing accuracy and reliability of the system.

[0005] This application provides a multi-sensor dynamic calibration method based on a mathematical model. The method includes: building a dynamic weighing transmission structure, which includes a mobile platform and a weighing unit; connecting the dynamic weighing transmission structure to obtain a historical weighing sample data set, which includes the historical weighing transmission data of the weighing unit and the historical movement control data of the mobile platform; training a dynamic disturbance filtering model according to the historical weighing sample data set and weighing error label samples, where the weighing error label is the error between the historical weighing transmission data and the actual weight data of the item. Among them, the dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is obtained by introducing the mean square error loss and performing supervised training on the historical weighing sample data set and the weighing error label samples; performing sensing integration on the target item through the weighing unit to obtain real-time weighing transmission data, sending the real-time weighing transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; compensating and calibrating the weighing result of the target item according to the weighing compensation data, and outputting the calibrated weighing transmission data.

[0006] In an implementation manner, the weighing unit includes a static weighing unit and a dynamic weighing unit. The static weighing unit is detachably disposed at the front end of the mobile platform, and the dynamic weighing unit is integrated within the mobile platform. The method includes: establishing a CAN bus communication link between the static weighing unit and the dynamic weighing unit and a central control module; placing an article sample on the static weighing unit, and storing the recorded static weighing transmission data as the actual weight data of the article in the central control module through the CAN bus communication link; placing the article sample on the dynamic weighing unit, and transmitting the recorded dynamic weighing transmission data to the central control module through the CAN bus communication link; obtaining a weighing error label sample of the article sample according to the static weighing transmission data and the dynamic weighing transmission data for training a dynamic disturbance filtering model.

[0007] In an implementation manner, the dynamic weighing unit includes a strain gauge sensor or a piezoelectric sensor. The number of the strain gauge sensors or the piezoelectric sensors is at least 2, and they are distributed and integrated within the mobile platform; and it is used to integrate the sensing data of the strain gauge sensor or the piezoelectric sensor as dynamic weighing transmission data through the CAN bus communication link.

[0008] In an implementation manner, the dynamic weighing unit includes a strain gauge sensor and a piezoelectric sensor. The total number of the strain gauge sensors and the piezoelectric sensors is at least 2, and they are distributed and integrated within the mobile platform; and it is used to integrate the sensing data of the strain gauge sensor and the piezoelectric sensor as dynamic weighing transmission data through the CAN bus communication link.

[0009] In an implementation manner, training a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label sample, the method includes: segmenting the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set into fixed-length time windows, and extracting the state fluctuation features of each window; wherein, the state fluctuation features include the acceleration, speed, signal fluctuation standard deviation and response delay of the current window; constructing a dynamic disturbance filtering model based on a Kalman filter, inputting the state fluctuation features of each window to calculate the error estimation value of each window, introducing the mean square error loss to iteratively train the weighing error label sample of the historical weighing sample data set and the error estimation value to obtain a dynamic disturbance filtering model.

[0010] In an implementation manner, the modeling of the Kalman filter includes setting state variables, defining a state transition equation and an observation equation according to the state variables; and according to the state transition equation and the observation equation, predicting and calculating the error estimation value of each window according to the state fluctuation features of each window.

[0011] In an implementation manner, after the training of the dynamic disturbance filtering model is completed, the static weighing unit is deactivated, and the dynamic weighing unit and the dynamic disturbance filtering model are activated; the target item is sensed and integrated by the activated dynamic weighing unit to obtain real-time dynamic weighing transmission data, and the weighing compensation data is obtained by invoking the dynamic disturbance filtering model.

[0012] In an implementation manner, when the dynamic weighing unit integrates multiple sensors of the same type, multiple dynamic weighing data corresponding to the multiple sensors of the same type are synchronously received; channel feature vectors of the multiple sensors of the same type are calculated according to the multiple dynamic weighing data, including the weighing average value, the standard deviation reflecting the fluctuation, and the difference between each channel; the multiple dynamic weighing data are fused according to the channel feature vectors, and the real-time weighing transmission data are output.

[0013] In an implementation manner, when the dynamic weighing unit integrates multiple sensors of different types, multiple dynamic weighing data corresponding to the multiple sensors of different types are synchronously received; multiple credibility weights of the multiple sensors of different types are calculated according to the historical stability of the multiple sensors of different types; the multiple dynamic weighing data are fused according to the multiple credibility weights, and the real-time weighing transmission data are output.

[0014] The present application further provides a multi-sensor dynamic calibration system based on a mathematical model, including: a weighing transmission structure building module for building a dynamic weighing transmission structure, where the dynamic weighing transmission structure includes a moving platform and a weighing unit; a data acquisition module for connecting to the dynamic weighing transmission structure to obtain a historical weighing sample data set, where the historical weighing sample data set includes the historical weighing transmission data of the weighing unit and the historical movement control data of the moving platform; a disturbance filtering module for training a dynamic disturbance filtering model according to the historical weighing sample data set and a weighing error label sample, where the weighing error label is the error between the historical weighing transmission data and the actual weight data of the item, and where the dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is supervised and trained based on the historical weighing sample data set and the weighing error label sample by introducing the mean square error loss; a weighing compensation module for sensing and integrating the target item through the weighing unit to obtain real-time weighing transmission data, and sending the real-time weighing transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; a weighing calibration module for compensating and calibrating the weighing result of the target item according to the weighing compensation data, and outputting the calibrated weighing transmission data.

[0015] A multi-sensor dynamic calibration method and system based on a mathematical model proposed in this application. By building a dynamic weighing transmission structure, the dynamic weighing transmission structure includes a moving platform and a weighing unit; connecting the dynamic weighing transmission structure to obtain a historical weighing sample data set, the historical weighing sample data set includes the historical weighing transmission data of the weighing unit and the historical movement control data of the moving platform; training a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label sample, the weighing error label is the error between the historical weighing transmission data and the actual weight data of the item, wherein the dynamic disturbance filtering model includes a Kalman filter, and is obtained by introducing a mean square error loss to supervise and train the Kalman filter based on the historical weighing sample data set and the weighing error label sample; performing sensing integration on the target item through the weighing unit to obtain real-time weighing transmission data, and sending the real-time weighing transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; compensating and calibrating the weighing result of the target item according to the weighing compensation data, and outputting the calibrated weighing transmission data. This solves the technical problems in the prior art that the weighing data is unstable and has a large error due to factors such as the movement of the item, the acceleration and vibration of the conveyor belt during the dynamic weighing process. It realizes high-precision dynamic weighing results that are stable and close to the real weight through multi-sensor fusion and disturbance compensation based on Kalman filtering in a dynamic environment, and improves the weighing accuracy and reliability. Description of the Drawings

[0016] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0017] Figure 1 Schematic flowchart of the multi-sensor dynamic calibration method based on a mathematical model provided by the embodiment of the present application; Figure 2 Schematic structural diagram of the multi-sensor dynamic calibration system based on a mathematical model provided by the embodiment of the present application; Figure 3 Schematic diagram of weighing compensation of the multi-sensor dynamic calibration method based on a mathematical model provided by the embodiment of the present application.

[0018] Description of the reference numerals: weighing transmission structure building module 11, data acquisition module 12, disturbance filtering module 13, weighing compensation module 14, weighing calibration module 15. Detailed implementation manners

[0019] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the specific implementation manners of the present application are specifically given below.

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0021] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first\second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0022] The embodiments of the present application provide a multi-sensor dynamic calibration method and system based on a mathematical model, as Figure 1 shown, the method includes: Step S100, building a dynamic weighing transmission structure, the dynamic weighing transmission structure includes a moving platform and a weighing unit; Step S200, connecting the dynamic weighing transmission structure, obtaining a historical weighing sample data set, the historical weighing sample data set includes the historical weighing transmission data of the weighing unit and the historical movement control data of the moving platform; Step S300, training a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label sample, the weighing error label is the error between the historical weighing transmission data and the actual weight data of the article, wherein, the dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is obtained by introducing the mean square error loss and performing supervised training on the historical weighing sample data set and the weighing error label sample.

[0023] Build a dynamic weighing and transmission structure, where the dynamic weighing and transmission structure includes a mobile platform and a weighing unit; the mobile platform is a transmission structure that can perform weighing functions. For example, in a packaging factory, the mobile platform can be an automated conveyor belt with an adjustable speed of up to 2 m / s to adapt to different production speed requirements. By placing items on the conveyor belt, the weighing unit on the transmission structure is used to weigh the items. However, due to the movement of the items and the influence of environmental parameters, the stability of weighing will be affected during this period. Therefore, it is necessary to eliminate the weighing error in subsequent processing. By connecting the dynamic weighing and transmission structure, a historical weighing sample data set during the historical operation process is obtained. The historical weighing sample data set includes the historical weighing transmission data of the weighing unit, the historical movement control data of the mobile platform, and the corresponding environmental parameters based on the time series. The environmental parameters include temperature data, humidity data, etc. The historical weighing transmission data is the weight reading of the items collected by the sensor. The historical movement control data is the control parameters of the conveyor belt, including relevant control data such as set speed, acceleration, and transportation slope. Further, using the obtained historical weighing sample data set and the weighing error label sample as training data, train a dynamic disturbance filtering model. The weighing error label is the error between the historical weighing transmission data and the actual weight data of the items. The dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is supervised and trained based on the historical weighing sample data set and the weighing error label sample by introducing the mean square error loss.

[0024] The method provided by the embodiment of the present application further includes: establishing a CAN bus communication link between the static weighing unit and the dynamic weighing unit and the central control module; placing an item sample on the static weighing unit, and storing the recorded static weighing transmission data as the actual weight data of the item in the central control module through the CAN bus communication link; placing the item sample on the dynamic weighing unit, and transmitting the recorded dynamic weighing transmission data to the central control module through the CAN bus communication link; according to the static weighing transmission data and the dynamic weighing transmission data, obtaining the weighing error label sample of the item sample for training the dynamic disturbance filtering model.

[0025] The weighing unit includes a static weighing unit and a dynamic weighing unit. The static weighing unit is detachably arranged at the front end of the mobile platform, and the dynamic weighing unit is integrated into the mobile platform. The method includes: arranging a static weighing unit such as a high-precision platform scale near the entrance of the conveyor belt, establishing a CAN bus communication link between the static weighing unit and the dynamic weighing unit and the central control module, and using shielded twisted pair to connect each sensor, the static platform scale and the central control module in series through the CAN bus communication link. Subsequently, place the item sample on the static weighing unit, and store the recorded static weighing transmission data as the actual weight data of the item in the central control module through the CAN bus communication link. Subsequently, place the item sample on the dynamic weighing unit, and transmit the recorded dynamic weighing transmission data to the central control module through the CAN bus communication link. Further, according to the static weighing transmission data and the dynamic weighing transmission data recorded by the central control module, by obtaining the sample data with errors between the static weighing transmission data and the dynamic weighing transmission data, obtain the weighing error label sample of the item sample. The weighing error label sample of the item sample includes the static weighing transmission data, the dynamic weighing transmission data, and the corresponding error parameter labels. The weighing error label sample of the item sample is used to train the dynamic disturbance filtering model.

[0026] The method provided by the embodiment of the present application further includes: the dynamic weighing unit includes a strain gauge sensor or a piezoelectric sensor, and the number of the strain gauge sensors or the piezoelectric sensors is at least 2 and is distributed and integrated into the mobile platform; and is used to integrate the sensing data of the strain gauge sensor or the piezoelectric sensor as the dynamic weighing transmission data through the CAN bus communication link.

[0027] The dynamic weighing unit includes a strain gauge sensor or a piezoelectric sensor, and the strain gauge sensor or the piezoelectric sensor is arranged at multiple positions in a uniformly distributed manner. The number of the strain gauge sensors or the piezoelectric sensors is at least 2 and is distributed and integrated into the mobile platform. And integrate the sensing data of the strain gauge sensor or the piezoelectric sensor as the dynamic weighing transmission data through the CAN bus communication link.

[0028] The method provided by the embodiment of the present application further includes: the dynamic weighing unit includes a strain gauge sensor and a piezoelectric sensor, and the number of the strain gauge sensors or the piezoelectric sensors is at least 2 and is distributed and integrated into the mobile platform; and is used to integrate the sensing data of the strain gauge sensor or the piezoelectric sensor as the dynamic weighing transmission data through the CAN bus communication link.

[0029] The dynamic weighing unit includes a strain sensor and a piezoelectric sensor. At this time, the types of sensors set in the dynamic weighing unit are different, and the strain sensor and the piezoelectric sensor are arranged in a uniform distribution at multiple positions. The number of the strain sensors or the piezoelectric sensors is at least 2, and they are distributed and integrated in the moving platform. And the sensing data of the strain sensor or the piezoelectric sensor is integrated through the CAN bus communication link as dynamic weighing transmission data.

[0030] The method provided by the embodiment of the present application further includes: segmenting the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set into fixed-length time windows, and extracting the state fluctuation characteristics of each window. Among them, the state fluctuation characteristics include the acceleration, speed, signal fluctuation standard deviation, and response delay of the current window; constructing a dynamic disturbance filtering model based on the Kalman filter, inputting the state fluctuation characteristics of each window to calculate the error estimation value of each window, introducing the mean square error loss, and iteratively training the weighing error label samples of the historical weighing sample data set and the error estimation value to obtain the dynamic disturbance filtering model.

[0031] Training the dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label samples, the method includes: segmenting the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set into fixed-length time windows. That is, the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set are divided into multiple data segments according to a fixed-duration window. For example, the fixed-length time window can be set to 10S, and the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set are divided into multiple data segments through the fixed-length time window. And extract the state fluctuation characteristics of each window. The state fluctuation characteristics include the acceleration, speed, signal fluctuation standard deviation, and response delay of the current window. Exemplarily, according to the collected historical samples, segmented according to a 10-second fixed time window, the state fluctuation characteristics of a window extracted are as follows: temperature 27.6°C, humidity 36%, average acceleration 0.15m / s 2 , average speed 1.0m / s, signal fluctuation standard deviation 5g, response delay time 50ms. The state fluctuation characteristics are used to measure the disturbance received by the data of each window. The acceleration is the average acceleration within the window, the speed is the average speed within the window, the signal fluctuation standard deviation is the standard deviation of the weighing data within the window, reflecting the jitter amplitude, and the response delay is the response time difference of the weighing unit to the movement instruction.

[0032] After feature extraction, a core modeling method based on the Kalman filter for dynamic disturbance filtering is used to construct a dynamic disturbance filtering model. Since the influence on weighing varies under different environmental parameters, when constructing the dynamic disturbance filtering model, the state fluctuation characteristics of the corresponding windows under different environmental parameters are input into the Kalman filter. Initially, the parameters of the Kalman filter are set, such as the state covariance matrix, the observation covariance matrix, etc. The predicted weighing error estimate value is calculated based on the current state characteristics of each window. Further, the model predicts an error estimate value for each time window, indicating the possible deviation of the weighing result of that window. The mean square error (MSE) loss function is introduced, and the error estimate value is compared with the weighing error label sample corresponding to the environmental parameters of that window in the historical weighing sample dataset, that is, the difference between the dynamic weighing data and the actual weight data of the item. By minimizing the mean square error, the parameters of the Kalman filter are iteratively optimized. The iteration continues until the model reaches stability and convergence on the validation set, and then a dynamic disturbance filtering model for real-time compensation is obtained. The dynamic disturbance filtering models for real-time compensation can be set to multiple according to different environmental parameters, and each dynamic disturbance filtering model for real-time compensation corresponds to processing the real-time weighing transmission data within a range of a certain environmental parameter interval.

[0033] The method provided by the embodiment of this application further includes: the modeling of the Kalman filter includes setting state variables, and defining a state transition equation and an observation equation according to the state variables; according to the state transition equation and the observation equation, the error estimate value of each window is predicted and calculated based on the state fluctuation characteristics of each window.

[0034] The modeling of the Kalman filter includes setting state variables, and the state variables are represented as X t = [w t ,g t ,v t ,a t ᵀ, where w t is the actual weight of the item, g t is the estimated weight of the weighing sensor, v t is the speed, and a t is the acceleration. According to the state variables, a state transition equation and an observation equation are defined. The form of the state transition equation is: X t+1 = A×X t + B×u t + c t , where A is the state transition matrix, B is the control input matrix, t represents the current moment, u t is the control input quantity, and c tis the process noise, and the state transition equation predicts the next state from the state at the current moment. The observation equation describes the correspondence between the actual observation data of the weighing sensor and the internal state of the system. The form of the observation equation is: Z t = H×X t + d t , where H is the observation matrix, and d t is the observation noise. The current state is predicted using the state transition equation, and the predicted state is adjusted by using the observation data, i.e., the current weighing signal and motion characteristics, for calibration through the observation equation. Furthermore, the error estimate value of the current window is output, i.e., the prediction of the weighing error. The error estimate value is compared with the historical weighing error label samples, and hyperparameters such as the noise covariance matrix in the Kalman filter are optimized by minimizing the mean square error. By continuously correcting the state transition matrix A, the observation matrix H, etc., the mean square error is finally reduced to less than 8% of the original error. Through the above steps, the Kalman filter continuously receives new data, updates its state estimate, and effectively extracts the errors caused by changes in the conveyor belt speed, acceleration, or other factors from the known dynamic weighing data. Thus, the influence of dynamic disturbances on the weighing result is filtered out in real time and accurately.

[0035] The method provided by the embodiment of the present application further includes: Step S400, performing sensing integration on the target item through the weighing unit to obtain real-time weighing transmission data, and sending the real-time weighing transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; Step S500, compensating and calibrating the weighing result of the target item according to the weighing compensation data, and outputting the calibrated weighing transmission data.

[0036] After the dynamic disturbance filtering model is trained, it enters the real-time application stage of the model. The weighing unit performs sensing integration on the target item to obtain real-time weighing transmission data. The real-time weighing transmission data also includes environmental data collected in the corresponding time window, i.e., average temperature data, humidity data, etc. The real-time weighing transmission data is segmented by a fixed-length time window, and then the segmented real-time weighing transmission data is sent to the dynamic disturbance filtering model corresponding to the current environmental parameters to obtain the weighing compensation data under the current environmental parameters. Finally, according to the obtained weighing compensation data, the weighing result of the target item is compensated and calibrated, that is, the average measured value in each fixed-length time window is increased or decreased by the weighing compensation data to complete the calibration compensation, and the calibrated weighing transmission data is output, as Figure 3 shown in the weighing compensation schematic diagram of the multi-sensor dynamic calibration method based on a mathematical model. Exemplarily, before calibration in a low-speed working condition (0.5 m / s), the average error is ±45 grams, and after calibration, the average error is ±3 grams, with an overall reduction amplitude of approximately 93%. Exemplarily, the data table of the dynamic weighing calibration case is shown in Table 1.

[0037] Table 1: Dynamic weighing calibration case data Number Sensor type Speed m / s Acceleration m / s² Standard deviation of signal fluctuation Response delay ms Error before calibration g Error after calibration g Percentage of error reduction 1 Strain gauge sensor ×4 0.5 0.1 50g 50 ±45 ±3 93% 2 Piezoelectric sensor ×2 1.2 0.2 80g 80 ±78 ±6 92% 3 Hybrid sensor (strain 2 + piezoelectric 2) 2 0.3 150g 150 ±120 ±8 93% 4 Strain gauge sensor ×6 0.8 0.15 60g 60 ±60 ±4 93% 5 Piezoelectric sensor ×4 1.5 0.25 120g 100 ±95 ±7 92.60% Thus, it is ensured that the final weighing result can reflect the actual static weight of the item as much as possible to offset the error caused by dynamic interference. Even when the conveyor belt is moving at high speed or there is disturbance due to mechanical vibration, the system can stably output the weight of the item close to the real value, greatly improving the accuracy and reliability of dynamic weighing.

[0038] The method provided by the embodiment of the present application further includes: performing sensing integration on the target item through the enabled dynamic weighing unit to obtain real-time dynamic weighing transmission data, and calling the dynamic disturbance filtering model to obtain weighing compensation data.

[0039] After the training of the dynamic disturbance filtering model is completed, at this time, the dynamic disturbance filtering model already has a certain error estimation ability. Then, the static weighing unit is deactivated, and the dynamic weighing unit and the dynamic disturbance filtering model are activated. Perform sensing integration on the target item through the enabled dynamic weighing unit to obtain real-time dynamic weighing transmission data. By inputting the real-time dynamic weighing transmission data into the dynamic disturbance filtering model according to a fixed time length window, obtain the weighing compensation data corresponding to each fixed time length window, and perform weighing data compensation.

[0040] The method provided by the embodiment of the present application further includes: calculating the channel feature vectors of the multiple same-type sensors according to the multiple dynamic weighing data, including the weighing average value, the standard deviation reflecting the fluctuation, and the difference between each channel; fusing the multiple dynamic weighing data according to the channel feature vectors, and outputting the real-time weighing transmission data.

[0041] When the dynamic weighing unit integrates multiple same-type sensors, at the same time node, synchronously receive the multiple dynamic weighing data corresponding to the multiple same-type sensors. Calculate the channel feature vectors of the multiple same-type sensors according to the multiple dynamic weighing data, and the channel feature vectors include the weighing average value, the standard deviation reflecting the fluctuation, and the difference between each channel. Among them, the weighing average value is the arithmetic average value of the sampling points of the channel within a fixed time length window, and the signal fluctuation standard deviation is the data standard deviation of the channel within a fixed time length window, reflecting the noise level. The difference between channels is the absolute difference between the actual measured value of each sensor of the channel within a fixed time length and the average value of all channels within a fixed time length. Finally, fuse the multiple dynamic weighing data according to the channel feature vectors. When performing channel feature vector fusion, perform weighted summation on the channel feature vectors obtained by each channel, and the weight is obtained based on the historical stability of the sensor, so as to output the real-time weighing transmission data.

[0042] The method provided by the embodiment of the present application further includes: calculating multiple credibility weights of the multiple different types of sensors according to the historical stability of the multiple different types of sensors; fusing the multiple dynamic weighing data according to the multiple credibility weights, and outputting the real-time weighing transmission data.

[0043] When the dynamic weighing unit integrates multiple different types of sensors, multiple dynamic weighing data corresponding to the multiple different types of sensors are synchronously received. According to the historical stability of the multiple different types of sensors, the calculation of the credibility weights of different types of sensors is carried out. The historical stability is the ratio of the measurement accurate parameters to all measurement parameters in the historical record. The credibility weight of each different type of sensor is the ratio of the historical stability of each different type of sensor to the sum of the historical stabilities of all different types of sensors. For example, the calculation result of the credibility weight of the strain gauge sensor is 0.7, and the calculation result of the credibility weight of the piezoelectric sensor is 0.3. Weighted summation is carried out according to the multiple credibility weights, so as to fuse the multiple dynamic weighing data and output the real-time weighing transmission data.

[0044] In the above, reference is made to Figure 1 The multi-sensor dynamic calibration method based on a mathematical model according to the embodiment of the present invention is described in detail. Next, reference will be made to Figure 2 Describe the multi-sensor dynamic calibration system based on a mathematical model according to the embodiment of the present invention.

[0045] The multi-sensor dynamic calibration system based on a mathematical model according to the embodiment of the present invention solves the technical problems of unstable weighing data and large errors caused by factors such as the movement of items, the acceleration and vibration of the conveyor belt during the dynamic weighing process in the prior art. The technical effect of outputting a high-precision dynamic weighing result that is stable and close to the real weight through multi-sensor fusion and disturbance compensation based on Kalman filtering in a dynamic environment is achieved, and the weighing accuracy and reliability are improved. The multi-sensor dynamic calibration system based on a mathematical model includes: a weighing transmission structure building module 11, a data acquisition module 12, a disturbance filtering module 13, a weighing compensation module 14, and a weighing calibration module 15.

[0046] The weighing and transmission structure building module 11 is used to build a dynamic weighing and transmission structure, and the dynamic weighing and transmission structure includes a moving platform and a weighing unit; the data acquisition module 12 is used to connect to the dynamic weighing and transmission structure to obtain a historical weighing sample data set, and the historical weighing sample data set includes the historical weighing and transmission data of the weighing unit and the historical movement control data of the moving platform; the disturbance filtering module 13 is used to train a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label sample, and the weighing error label is the error between the historical weighing and transmission data and the actual weight data of the item. Among them, the dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is supervised and trained based on the historical weighing sample data set and the weighing error label sample by introducing the mean square error loss; the weighing compensation module 14 is used to perform sensing integration on the target item through the weighing unit, obtain real-time weighing and transmission data, and send the real-time weighing and transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; the weighing calibration module 15 is used to compensate and calibrate the weighing result of the target item according to the weighing compensation data, and output the calibrated weighing and transmission data.

[0047] Next, the specific configuration of the weighing and transmission structure building module 11 will be described in detail. The weighing and transmission structure building module 11 may further include: the weighing unit includes a static weighing unit and a dynamic weighing unit, the static weighing unit is detachably arranged at the front end of the moving platform, and the dynamic weighing unit is integrated in the moving platform. The method includes: establishing a CAN bus communication link between the static weighing unit and the dynamic weighing unit and the central control module; placing an item sample on the static weighing unit, and storing the recorded static weighing and transmission data as the actual weight data of the item in the central control module through the CAN bus communication link; placing the item sample on the dynamic weighing unit, and transmitting the recorded dynamic weighing and transmission data to the central control module through the CAN bus communication link; obtaining the weighing error label sample of the item sample according to the static weighing and transmission data and the dynamic weighing and transmission data, for training the dynamic disturbance filtering model.

[0048] Next, the specific configuration of the weighing and transmission structure building module 11 will be further described in detail. The weighing and transmission structure building module 11 further includes: the dynamic weighing unit includes a strain gauge sensor or a piezoelectric sensor, the number of the strain gauge sensor or the piezoelectric sensor is at least 2, and they are distributed and integrated in the moving platform; for integrating the sensing data of the strain gauge sensor or the piezoelectric sensor as the dynamic weighing and transmission data through the CAN bus communication link.

[0049] Next, the specific configuration of the weighing and transmission structure building module 11 will be further described in detail. The weighing and transmission structure building module 11 further includes: The dynamic weighing unit includes a strain sensor and a piezoelectric sensor. The number of the strain sensors or the piezoelectric sensors is at least 2, and they are distributed and integrated in the moving platform; and is used to integrate the sensing data of the strain sensors or the piezoelectric sensors through the CAN bus communication link as dynamic weighing transmission data.

[0050] Next, the specific configuration of the disturbance filtering module 13 will be described in detail. The disturbance filtering module 13 further includes: training a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label samples. The method includes: segmenting the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set into fixed-length time windows, and extracting the state fluctuation features of each window; wherein, the state fluctuation features include the acceleration, speed, signal fluctuation standard deviation, and response delay of the current window; constructing a dynamic disturbance filtering model based on the Kalman filter, inputting the state fluctuation features of each window to calculate the error estimation value of each window, introducing the mean square error loss to iteratively train the weighing error label samples of the historical weighing sample data set and the error estimation value, and obtaining the dynamic disturbance filtering model.

[0051] Next, the specific configuration of the disturbance filtering module 13 will be described in detail. The disturbance filtering module 13 further includes: The modeling of the Kalman filter includes setting state variables, and defining a state transition equation and an observation equation according to the state variables; according to the state transition equation and the observation equation, predicting and calculating the error estimation value of each window according to the state fluctuation features of each window.

[0052] Next, the specific configuration of the weighing compensation module 14 will be described in detail. The weighing compensation module 14 further includes: after the training of the dynamic disturbance filtering model is completed, deactivating the static weighing unit, and activating the dynamic weighing unit and the dynamic disturbance filtering model; sensing and integrating the target item through the activated dynamic weighing unit to obtain real-time dynamic weighing transmission data, and calling the dynamic disturbance filtering model to obtain weighing compensation data.

[0053] Next, the specific configuration of the weighing compensation module 14 will be described in detail. The weighing compensation module 14 further includes: when the dynamic weighing unit integrates multiple sensors of the same type, synchronously receiving multiple dynamic weighing data corresponding to the multiple sensors of the same type; calculating the channel feature vectors of the multiple sensors of the same type according to the multiple dynamic weighing data, including the weighing average value, the standard deviation of the reaction fluctuation, and the difference between each channel; fusing the multiple dynamic weighing data according to the channel feature vectors, and outputting the real-time weighing transmission data.

[0054] Next, the specific configuration of the weighing compensation module 14 will be described in detail. The weighing compensation module 14 further includes: when the dynamic weighing unit integrates multiple different types of sensors, synchronously receiving multiple dynamic weighing data corresponding to the multiple different types of sensors; calculating multiple credibility weights of the multiple different types of sensors according to the historical stability of the multiple different types of sensors; and fusing the multiple dynamic weighing data according to the multiple credibility weights to output the real-time weighing transmission data.

[0055] The multi-sensor dynamic calibration system based on a mathematical model provided by the embodiments of the present invention can execute the multi-sensor dynamic calibration method based on a mathematical model provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution of the method.

[0056] Although this application makes various references to certain modules in the system according to the embodiments of this application, however, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for easy differentiation from each other and do not limit the protection scope of the present invention.

[0057] The above specific implementation manners do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A multi-sensor dynamic calibration method based on a mathematical model, characterized in that, The method includes: Construct a dynamic weighing and transmission structure, which includes a moving platform and a weighing unit; Connect the dynamic weighing and transmission structure to obtain a historical weighing sample data set, which includes the historical weighing transmission data of the weighing unit and the historical movement control data of the moving platform; Train a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label samples. The weighing error label is the error between the historical weighing transmission data and the actual weight data of the item. Among them, the dynamic disturbance filtering model includes a Kalman filter, and is obtained by introducing the mean square error loss and supervising the training of the Kalman filter based on the historical weighing sample data set and the weighing error label samples; Perform sensing integration on the target item through the weighing unit to obtain real-time weighing transmission data, and send the real-time weighing transmission data into the dynamic disturbance filtering model to obtain weighing compensation data; Compensate and calibrate the weighing result of the target item according to the weighing compensation data, and output the calibrated weighing transmission data.

2. The method according to claim 1, characterized in that, The weighing unit includes a static weighing unit and a dynamic weighing unit. The static weighing unit is detachably arranged at the front end of the moving platform, and the dynamic weighing unit is integrated in the moving platform. The method includes: Establish a CAN bus communication link between the static weighing unit and the dynamic weighing unit and the central control module; Place the item sample on the static weighing unit, and store the recorded static weighing transmission data as the actual weight data of the item in the central control module through the CAN bus communication link; Place the item sample on the dynamic weighing unit, and transmit the recorded dynamic weighing transmission data to the central control module through the CAN bus communication link; Obtain the weighing error label samples of the item sample according to the static weighing transmission data and the dynamic weighing transmission data, which are used to train the dynamic disturbance filtering model.

3. The method according to claim 2, characterized in that, The dynamic weighing unit includes a strain sensor or a piezoelectric sensor. The number of the strain sensor or the piezoelectric sensor is at least 2, and they are distributed and integrated in the moving platform; It is used to integrate the sensing data of the strain sensor or the piezoelectric sensor as the dynamic weighing transmission data through the CAN bus communication link.

4. The method according to claim 2, characterized in that, The dynamic weighing unit includes a strain sensor and a piezoelectric sensor. The total number of the strain sensor and the piezoelectric sensor is at least 2, and they are distributed and integrated in the moving platform; It is used to integrate the sensing data of the strain sensor and the piezoelectric sensor as the dynamic weighing transmission data through the CAN bus communication link.

5. The method according to claim 1, characterized in that, Train a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label samples. The method includes: Segment the dynamic weighing transmission data and the historical movement control data in the historical weighing sample data set into fixed-length time windows, and extract the state fluctuation characteristics of each window; Among them, the state fluctuation characteristics include the acceleration, speed, standard deviation of signal fluctuation, and response delay of the current window; Construct a dynamic disturbance filtering model based on the Kalman filter, input the state fluctuation characteristics of each window to calculate the error estimation value of each window, and introduce the mean square error loss to iteratively train the weighing error label samples of the historical weighing sample data set and the error estimation value to obtain the dynamic disturbance filtering model.

6. The method according to claim 5, characterized in that, The modeling of the Kalman filter includes setting state variables and defining a state transition equation and an observation equation according to the state variables; According to the state transition equation and the observation equation, predict and calculate the error estimation value of each window based on the state fluctuation characteristics of each window.

7. The method according to claim 2, characterized in that, After the training of the dynamic disturbance filtering model is completed, deactivate the static weighing unit and activate the dynamic weighing unit and the dynamic disturbance filtering model; Sense and integrate the target item through the activated dynamic weighing unit to obtain real-time dynamic weighing transmission data, and call the dynamic disturbance filtering model to obtain weighing compensation data.

8. The method according to claim 3, characterized in that, When the dynamic weighing unit integrates multiple sensors of the same type, synchronously receive multiple dynamic weighing data corresponding to the multiple sensors of the same type; Calculate the channel feature vectors of the multiple sensors of the same type according to the multiple dynamic weighing data, including the weighing average value, the standard deviation of the reaction fluctuation, and the difference between each channel; Fuse the multiple dynamic weighing data according to the channel feature vectors and output the real-time weighing transmission data.

9. The method according to claim 4, characterized in that, When the dynamic weighing unit integrates multiple sensors of different types, synchronously receive multiple dynamic weighing data corresponding to the multiple sensors of different types; Calculate multiple credibility weights of the multiple sensors of different types according to the historical stability of the multiple sensors of different types; Fuse the multiple dynamic weighing data according to the multiple credibility weights and output the real-time weighing transmission data.

10. A multi-sensor dynamic calibration system based on a mathematical model, characterized in that, The system is used to execute the method according to any one of claims 1-9, and the system includes: A weighing transmission structure building module for building a dynamic weighing transmission structure, where the dynamic weighing transmission structure includes a moving platform and a weighing unit; A data acquisition module for connecting to the dynamic weighing transmission structure to obtain a historical weighing sample data set, where the historical weighing sample data set includes the historical weighing transmission data of the weighing unit and the historical movement control data of the moving platform; A disturbance filtering module for training a dynamic disturbance filtering model according to the historical weighing sample data set and the weighing error label samples, where the weighing error label is the error between the historical weighing transmission data and the actual weight data of the item. Among them, the dynamic disturbance filtering model includes a Kalman filter, and the Kalman filter is obtained by supervised training based on the historical weighing sample data set and the weighing error label samples by introducing the mean square error loss; A weighing compensation module for sensing and integrating the target item through the weighing unit to obtain real-time weighing transmission data, and sending the real-time weighing transmission data to the dynamic disturbance filtering model to obtain weighing compensation data; A weighing calibration module, configured to compensate and calibrate the weighing result of the target item according to the weighing compensation data, and output the calibrated weighing transmission data.

Images