Long-term stable and reliable high-precision belt weigher based on artificial intelligence

Abstract

The invention discloses a long-term stable and reliable high-precision belt scale based on artificial intelligence, particularly relates to the field of electronic belt scales, and comprises an environment sensing module, a data acquisition module, a multi-modal data fusion module, a self-learning calibration module, an intelligent fault early warning module and a dynamic control module. According to the long-term stable and reliable high-precision belt weigher based on artificial intelligence, full-dimensional data acquisition is realized through a multi-sensor cooperative monitoring technology of an environment sensing module, the acquisition precision of environmental parameters is improved, and the metering stability in a complex environment is improved; through transfer learning of the self-learning calibration module and the convolutional neural network technology, the dynamic compensation parameter updating frequency is improved, and the metering precision attenuation rate in the whole life cycle is reduced; the weight is driven by the servo motor to be automatically loaded, updating of the self-learning calibration model is triggered in real time, the frequency of manual intervention and the maintenance cost are reduced, and the shutdown requirement of traditional chain code verification is avoided.

Description

Technical Field

[0001] This invention relates to the field of electronic belt scale technology, and more specifically, to a high-precision belt scale based on artificial intelligence that is stable and reliable in the long term. Background Technology

[0002] With the rapid development of industrial automation and intelligence, belt scales, as an important metering device, are widely used for weight monitoring and measurement in the material conveying process. Traditional belt scales mainly rely on mechanical structures and simple electronic sensors for weight measurement. They detect belt tension through weighing sensors and calculate material flow rate by combining speed sensors. However, traditional belt scales have long faced pain points such as accuracy being easily affected by environmental interference, maintenance relying on manual experience, and delayed fault response.

[0003] Currently, with the support of AI and IoT technologies, new momentum has been injected into the digital transformation of industrial enterprises, and some improvement methods have emerged to enhance the measurement performance of belt scales, including but not limited to increasing the number of idlers and sensor layout to improve the measurement coverage, and adopting matrix weighing units combined with AI algorithms to capture material distribution information in real time to reduce errors.

[0004] However, in actual use, it still has some shortcomings. For example, under harsh working conditions such as high temperature and dust, the sensor signal is easily interfered with, and the existing belt scale is difficult to achieve continuous and stable high-precision measurement. Continuous use is susceptible to the effects of temperature drift and mechanical wear, which leads to the decay of key performance indicators of long-term measurement. The calibration and maintenance of traditional belt scales require shutdown inspection and a lot of manual intervention. At the same time, it is impossible to identify hidden faults such as sensor aging and idler roller jamming in advance by judging anomalies only through hardware redundancy, which increases maintenance costs. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a high-precision belt scale based on artificial intelligence that is stable and reliable in the long term, and solves the problems mentioned in the background art through the following solution.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A high-precision belt scale based on artificial intelligence that offers long-term stability and reliability includes: Environmental sensing module: used to monitor the operating environment data of the target belt scale in real time through multiple types of high-precision acquisition devices; Data acquisition module: Based on the high-precision acquisition device, acquire the first measurement data of the target belt scale; Multimodal data fusion module: used to perform deep fusion processing on the first measurement data and the operating environment data through a multimodal data fusion algorithm to obtain the first measurement feature corresponding to the first measurement data; Self-learning calibration module: used to acquire a self-learning calibration model, input the first metrological feature into the self-learning calibration model for self-learning and self-calibration, and acquire the second metrological data; Intelligent fault early warning module: used to perform real-time analysis and monitoring of the second metering data through pre-built fault early warning rules, and obtain the second metering characteristics of the target belt scale; Dynamic control module: Based on the second metering feature, dynamically adjust the target belt scale.

[0007] Preferably, the environmental sensing module includes environmental temperature and humidity values ​​collected by a temperature and humidity sensor, acceleration, displacement and velocity of mechanical vibration in various directions collected by a three-dimensional vibration sensor, dynamic data of dust concentration collected by a laser dust sensor, and electromagnetic interference signals collected by an electromagnetic interference detection device.

[0008] Preferably, the data acquisition module receives the following first measurement data from the target belt scale: dynamic weight signal, speed fluctuation component, and belt tension.

[0009] Preferably, the self-learning calibration module acquires the self-learning calibration model, specifically including: Construct a dataset containing historical mapping relationships between environmental parameters and corresponding measurement deviations; Historical spatiotemporal features of the historical mapping relationship dataset are extracted using a convolutional neural network. A transfer learning algorithm is used to iteratively train a deep neural network model using the historical spatiotemporal features to obtain a self-learning calibration model.

[0010] Preferably, the self-learning calibration module acquires the second measurement data, specifically including: A benchmark measurement curve is generated based on the first measurement feature; Compensation parameters are generated by combining a self-learning calibration model; The compensation parameters are spatiotemporally aligned with the first measurement data, and calibration data is generated through a weighted fusion algorithm. The calibration data is output after confidence verification.

[0011] Preferably, the self-learning calibration module calculates the time based on the benchmark metrology curve corresponding to the first metrology feature. Compensation parameters at time Specifically, it is expressed as: in, Represented as in time The measurement deviation between the actual measured value and the reference measured value Represented as the baseline measurement curve, Represented as the current time index, This is represented by the preset environmental compensation coefficient. This is represented as the preset trend tracking coefficient.

[0012] Preferably, the intelligent fault early warning module pre-constructs fault early warning rules, specifically including: Data fluctuation threshold rules based on statistical process control; Wear and degradation prediction trend based on the cumulative operating time of the target belt scale; The correlation rules between temperature gradient anomalies and mechanical deformation; The data fluctuation threshold of statistical process control is the dynamic error data generated by comparing the theoretical value and the measured value of the weights, which are automatically loaded into the weighing area of ​​the target belt scale by a servo motor.

[0013] Preferably, the intelligent fault early warning module, which uses a servo motor to drive the automatic loading of weights, specifically includes: The preset permutation and combination modes include single-point loading, multi-point symmetrical loading, and dynamic moving loading. Based on the coupling relationship between the preset permutation and combination pattern and the belt speed, the theoretical weight distribution curve is calculated to dynamically correct the self-learning calibration model.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention utilizes multi-sensor collaborative monitoring technology of the environmental sensing module, integrating temperature, humidity, vibration, dust, and image sensors, combined with a laser Doppler velocimeter and an ultrasonic tension sensor, to achieve full-dimensional data acquisition, improve the accuracy of environmental parameter acquisition, and enhance measurement stability in complex environments; 2. This invention improves the update frequency of dynamic compensation parameters and reduces the rate of measurement accuracy decay over the entire life cycle by using transfer learning and convolutional neural network technology in the self-learning calibration module, thus improving the efficiency compared to traditional manual calibration. 3. This invention uses a servo motor to drive the automatic loading of weights, which are loaded into the weighing area according to a preset arrangement and combination pattern. This triggers the update of the self-learning calibration model in real time, reducing the frequency of manual intervention and maintenance costs, and avoiding the downtime requirements of traditional chain code verification. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a high-precision belt scale based on artificial intelligence that is stable and reliable in the long term, according to an embodiment of this application.

[0016] Figure 2This is a flowchart illustrating the steps of a multimodal data fusion module in a high-precision belt scale based on artificial intelligence, which obtains a first metrological feature according to an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0019] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0020] Example 1, as shown in the appendix Figure 1 The high-precision belt scale shown is based on artificial intelligence and is stable and reliable in the long term. It includes an environmental perception module, a data acquisition module, a multimodal data fusion module, a self-learning calibration module, an intelligent fault early warning module, and a dynamic control module.

[0021] It should be noted that the environmental sensing module is used to monitor the operating environment data of the target belt scale in real time through multiple types of high-precision acquisition devices; The data acquisition module collects the first measurement data of the target belt scale based on the high-precision acquisition device; The multimodal data fusion module is used to perform deep fusion processing on the first measurement data and the operating environment data through a multimodal data fusion algorithm to obtain the first measurement feature corresponding to the first measurement data; The self-learning calibration module is used to obtain a self-learning calibration model. The first metrological feature is input into the self-learning calibration model to perform self-learning and self-calibration, and then the second metrological data is obtained. The intelligent fault early warning module is used to perform real-time analysis and monitoring of the second metering data through pre-built fault early warning rules, and to obtain the second metering characteristics of the target belt scale; The dynamic control module dynamically adjusts the target belt scale based on the second metering feature.

[0022] Example 2, see appendix Figure 1 This embodiment is based on the above embodiment. In the environmental perception module, various types of high-precision sensors are pre-deployed to collect the operating environment data of the target belt scale. The various types of high-precision sensors pre-deployed include, but are not limited to, digital temperature and humidity sensors, vibration sensors, dust sensors, electromagnetic interference detection equipment, strain gauge load cells, speedometers, tension sensors, etc. The operating environment data of the target belt scale includes the ambient temperature and humidity values ​​collected by the temperature and humidity sensors, the acceleration, displacement and velocity of mechanical vibration in various directions collected by the three-dimensional vibration sensors, the dynamic data of dust concentration collected by the laser dust sensor, and the electromagnetic interference signals collected by the electromagnetic interference detection equipment.

[0023] In one possible implementation, real-time monitoring of the target belt scale's operating environment data includes: S101: By pre-deploying 6 sets of SHT35 digital temperature and humidity sensor units in the longitudinal axis direction of the target belt scale, time-series temperature and humidity data with an accuracy of ±0.2°C temperature deviation and ±1.5%RH humidity deviation are collected in real time synchronously. It should be noted that each set of digital temperature and humidity sensor units is arranged at the preset monitoring point according to the principle of equal spacing, and all sensor units at the preset monitoring point are encapsulated in a metal shell with an IP67 protection rating and integrated with a ceramic heat insulation base at the bottom. S102: By pre-deploying multiple types of vibration sensors at key mechanical nodes of the target belt scale, synchronous monitoring data of the acceleration, displacement, and velocity parameters of the mechanical vibration signal of the target belt scale in the XYZ orthogonal directions is achieved. It should be noted that the multiple types of vibration monitoring devices include, but are not limited to, X / Y / Z triaxial piezoelectric vibration sensor arrays, tilt-compensated vibration sensors, and wireless vibration nodes. Among them, the X / Y / Z triaxial piezoelectric vibration sensor array is pre-deployed in the motor base area to achieve gear meshing vibration in the full frequency band of 50-800Hz and acquire instantaneous acceleration and displacement of 0.01 mm within the ±50g range. The tilt-compensated vibration sensor is pre-installed symmetrically on both sides of the load-bearing beam to dynamically acquire vertical vibration velocity data of 0-1 m / s with an accuracy of ±0.5%FS. The wireless vibration node achieves three-dimensional vibration spectrum data at a sampling rate of 5kHz at the return idler roller. S103: A triangular monitoring matrix constructed by three sets of PM3006 laser scattering dust sensors pre-deployed at the feed end, weighing area, and discharge end of the target belt scale, dynamically collects data from 0 to 1000 μg / m³ in real time. 3 Dust concentration data within the measurement range; it should be noted that the laser scattering dust sensor has a built-in cyclone separator preprocessing unit; S104: A broadband electromagnetic radiation analyzer is pre-deployed in areas such as the electrical control cabinet and load cell cable tray of the target belt scale to acquire electromagnetic interference signals in the full frequency band from 1Hz to 60GHz in real time. It should be noted that the broadband electromagnetic radiation analyzer integrates a spectrum scanning function, performs a fast scan of the frequency band from 0.1 to 100MHz every 60 seconds, and uses an ADC module with a sampling rate of 200MS / s to capture transient pulses of <10nsEFT burst interference.

[0024] Example 3, see appendix Figure 1 This embodiment is based on the above embodiment. In the data acquisition module, based on the various types of high-precision sensors pre-deployed in the environmental perception module, the first measurement data of the target belt scale is collected. The first measurement data includes dynamic weight signal, speed fluctuation component and belt tension.

[0025] In one possible implementation, collecting the first measurement data of the target belt scale includes: S201: Eight sets of HBMPW15A high-precision strain gauge load cell nodes are pre-deployed along the axis of the load-bearing beam of the target belt scale based on the optimization results of finite element simulation. Simultaneously, the strain gauge load cells are connected to the load-bearing beam via universal bearings and equipped with bidirectional tie rod limiting devices to acquire dynamic weight signals. These dynamic weight signals are effective load data with a signal-to-noise ratio ≥80dB. It should be noted that the strain gauge load cell nodes include three sets of full-bridge sensors symmetrically installed 0.3m below the main load-bearing beam, covering a 2.5m effective weighing area of ​​the belt, and one set of temperature self-compensating sensors with built-in PT1000 platinum resistance thermometers deployed on each side of the weighing roller. It should be noted that the explanation of the finite element simulation optimization of sensor layout (such as stress distribution cloud map) is included.

[0026] S202: The belt running speed data is obtained by using a laser Doppler velocimeter pre-installed on the belt drive drum side of the target belt scale and an incremental photoelectric encoder configured synchronously. It should be noted that the laser Doppler velocimeter and the incremental photoelectric encoder perform FFT analysis on the acquired raw pulse signal and extract the wavelet packet energy entropy of the 0-50Hz characteristic frequency band to obtain the speed fluctuation component. S203: The belt tension is obtained by pre-installing four sets of ultrasonic tension sensors in the belt tensioning section of the target belt scale. It should be noted that the ultrasonic tension sensor configuration points include two sets of static tension baselines arranged in a V-shape near the drive end and two sets of dynamic tension fluctuations detected by horizontally symmetrically installed on the redirecting roller side.

[0027] Example 4, see appendix Figure 1 and attached Figure 2 This embodiment is based on the above embodiment. In the multimodal data fusion module, the first measurement data and the operating environment data are deeply fused to obtain the first measurement feature. The first measurement feature includes the original data information of the first measurement data and the operating environment data, as well as the derived features generated by the deep fusion process. The derived features include mechanical vibration spectrum features, power frequency interference intensity, load distribution feature matrix, vibration-weight error compensation coefficient, dust-velocity coupling factor, instantaneous flow data, and temperature and humidity gradient values.

[0028] It should be noted that deep fusion processing includes the following steps: S301: Perform spatiotemporal alignment and standardization on the primary measurement data and operating environment data; S301-1: Establish a unified spatiotemporal coordinate system: Using the laser Doppler velocimeter and incremental photoelectric encoder pre-installed in S202 as the synchronization reference, the first rising edge pulse output by the laser Doppler velocimeter is taken as the starting point to synchronously trigger the incremental photoelectric encoder counter and the data acquisition of all sensors, and to achieve μs-level timestamp alignment with the help of hardware synchronization circuit; the belt of the target belt scale is divided into N equidistant spatial grids along the running direction, and the physical positions of each pre-deployed sensor are mapped to the corresponding spatial grids; S301-2: Spatial Alignment: Spatial alignment of non-uniform sampled data is performed using a cubic spline interpolation algorithm according to the spatial grid order. Non-uniform sampled data includes, but is not limited to, vibration data at different belt positions in the operating environment data, electromagnetic interference signals that occur suddenly in a specific frequency band, and dust concentration in the first measurement data. S301-3: Standardization process: unify the dimensions of the primary measurement data and the operating environment data, and generate a spatiotemporally labeled dataset; S302: Extracting key features through multiphysics joint analysis; Furthermore, a short-time Fourier transform is performed on the mechanical vibrations collected by the three-dimensional vibration sensor in the operating environment data to extract the energy proportion in the 0-100Hz frequency band as the spectral characteristics of the mechanical vibration; a 50Hz bandpass filter is performed on the electromagnetic interference signals collected by the electromagnetic interference detection equipment in the operating environment data, and the root mean square value is calculated to obtain the power frequency interference intensity. Furthermore, based on the belt tension and dynamic weight signals in the first measurement data, a two-dimensional distribution matrix is ​​constructed to obtain the load distribution feature matrix; S303: Dynamic weight allocation based on entropy weight method; Furthermore, the key features extracted by S302 are normalized and transformed into dimensionless data in the 0-1 interval; all samples of each key feature are traversed to calculate the dispersion of the data distribution of each key feature; at the same time, the weight of each key feature is dynamically allocated according to the magnitude of information entropy. The greater the information entropy, the more information the feature contains, and the greater its weight. S304: Generate highly robust measurement characteristics; Furthermore, based on the weights allocated in S303, each key feature is weighted and fused to generate highly robust metrological features. These highly robust metrological features include, but are not limited to, vibration-weight error compensation coefficient, dust-velocity coupling factor, instantaneous flow data, and temperature and humidity gradient values.

[0029] Example 5, see appendix Figure 1 This embodiment, based on the above embodiment, obtains a self-learning calibration model in the self-learning calibration module, specifically including: S401: Construct a dataset containing historical mapping relationships of environmental parameters and corresponding measurement deviations; Specifically, temperature and humidity gradients, mechanical vibration spectrum characteristics, dust-velocity coupling factors, and power frequency interference intensity over a period of more than two years are obtained from 500 pre-deployed belt scales. The measured weights of each belt scale are compared with the offline static weighing results to calculate the relative measurement deviation. ;right Abnormal operating condition data is removed, and the remaining data is normalized using Min-Max, ultimately generating a dimension of... The historical mapping relationship dataset, in which, Represented as the number of samples, Represented as a time step, Represented as the number of data channels; S402: Extract the historical spatiotemporal features of the historical mapping relationship dataset using a convolutional neural network; Specifically, the constructed historical mapping relationship dataset is input into a convolutional neural network, which consists of multiple convolutional layers, pooling layers, and fully connected layers. Different convolutional kernels are used to perform sliding convolution operations on the historical mapping relationship dataset to extract and sample local features. After processing by multiple convolutional layers and pooling layers, the historical spatiotemporal features corresponding to the historical mapping relationship dataset are obtained. S403: Employ the transfer learning algorithm to iteratively train the deep neural network model using the historical spatiotemporal features to obtain a self-learning calibration model; Specifically, ResNet-34 pre-trained on ImageNet is used as the feature extractor. The parameters of the feature extractor are continuously adjusted by backpropagation algorithm using historical spatiotemporal features. When the performance of the feature extractor reaches the optimal level, the deep neural network model corresponding to the feature extractor is saved as a self-learning calibration model.

[0030] Furthermore, the first metrological feature transmitted from the multimodal data fusion module is input into the self-learning calibration model for self-learning and self-calibration to obtain the second metrological data, specifically including: S501: Generate a benchmark measurement curve based on the first measurement feature; Specifically, the first metrological feature output by the multimodal data fusion module is input into the self-learning calibration model to generate a benchmark metrological curve. The benchmark metrological curve is the metrological change trend of the target belt scale under normal conditions. Normal conditions are ideal conditions where the environment is stable and the target belt scale is fault-free. S502: Generate compensation parameters by combining a self-learning calibration model; In one possible approach, the time is calculated based on the benchmark measurement curve corresponding to the first measurement characteristic. Compensation parameters at time Specifically, it is expressed as: in, Represented as in time The measurement deviation between the actual measured value and the reference measured value Represented as the baseline measurement curve, Represented as the current time index, This represents the preset environmental compensation coefficient, with a value of 0.7. This represents the preset trend-following coefficient, with a value of 0.3. It should be noted that the benchmark measurement curve represents the measurement change trend of the target belt scale under normal operating conditions through the self-learning calibration model, specifically expressed as follows: in, Represented as a self-learning calibration model, This is represented by the value of the first measurement feature at historical time step T. Represented as an index for a historical time step. Represented as the current time index, The index representing the first measurement feature; S503: The compensation parameters are spatiotemporally aligned with the first measurement data, and calibration data is generated through a weighted fusion algorithm; Specifically, the compensation parameters The compensation parameters and the first measurement data are arranged chronologically and aligned with the positions corresponding to the timestamps of each second. If there is no directly corresponding first measurement data at a timestamp, linear interpolation is performed using adjacent known data. The compensation parameters and the first measurement data are converted into time series data with 1-second intervals. Based on the position code of the target belt scale, the compensation parameters are mapped to the corresponding weighing segments. After time and space alignment, the compensation parameters and the first measurement data are used to generate calibration data through a weighted fusion algorithm. ; S504: Output calibration data, i.e., second measurement data, after passing confidence verification; Specifically, if ,in, If expressed as the historical residual standard deviation, then S502 will be reimplemented.

[0031] Example 6, see appendix Figure 1 This embodiment is based on the above embodiment. In the intelligent fault early warning module, the second measurement data is compared and analyzed according to the pre-built fault early warning rules to obtain the second measurement features corresponding to the second measurement data. The second measurement features include the second measurement data and the derived features corresponding to the second measurement data. At the same time, the self-learning calibration model is triggered to iteratively update. Among them, the derived features corresponding to the second measurement data include the calibrated mechanical vibration spectrum features, power frequency interference intensity, load distribution feature matrix, vibration-weight error compensation coefficient, dust-velocity coupling factor, instantaneous flow data, and temperature and humidity gradient values.

[0032] It should be noted that the pre-constructed fault warning rules include: data fluctuation threshold rules based on statistical process control; wear and degradation prediction trend based on the cumulative operating time of the target belt scale; and correlation rules between abnormal temperature gradient and mechanical deformation. Among them, the data fluctuation threshold of statistical process control is dynamic error data generated by comparing the theoretical value and the measured value of the weights through automatic loading of weights into the weighing area of ​​the target belt scale by a servo motor.

[0033] Furthermore, the automatic loading of the weights by the servo motor automatically loads the standard weights into the weighing area of ​​the belt scale according to the preset program; based on the relative difference between the theoretical and measured weight values ​​of the weights, the error distribution matrix is ​​calculated; if the relative difference exceeds the error threshold three times in a row, an update command is sent to the self-learning calibration module to trigger the iterative optimization of the compensation parameters. In this embodiment, the error threshold is set to 0.5% of the theoretical weight value of the weights.

[0034] Furthermore, the control method for automatic loading of weights driven by a servo motor includes: using preset permutation and combination modes, which include single-point loading, multi-point symmetrical loading, and dynamic moving loading; and calculating the theoretical weight distribution curve based on the coupling relationship between the preset permutation and combination modes and the belt speed, so as to dynamically correct the self-learning calibration model.

[0035] It should be noted that the theoretical weight distribution curve is calculated based on the coupling relationship between the preset permutation and combination pattern and the belt speed. Specifically, it is expressed as: in, This represents the total number of preset permutation and combination patterns. Indicates an index representing a preset permutation and combination pattern. Represented as permutation and combination coefficients, Represented as the first The weight of a standard weight This is expressed as the real-time belt speed.

[0036] In one possible implementation, single-point loading in the preset permutation and combination mode involves driving a single standard weight along the belt running direction to a preset weighing area via a servo motor, staying for 10 seconds, and then returning to the initial position; multi-point symmetrical loading involves synchronously driving two sets of weights symmetrically to be loaded on both sides of the weighing area, with the positions on both sides ±0.5m from the center point, each set of weights weighing 10kg, and the loading time interval being 5 seconds; dynamic moving loading involves driving the weights to move continuously along the belt running direction at a speed of 0.2m / s to simulate a dynamic material distribution scenario.

[0037] In one possible implementation, dynamically correcting the self-learning calibration model includes: updating the current weight arrangement pattern, belt speed, and theoretical weight distribution curve to the historical mapping dataset; and dynamically adjusting the compensation parameters of the self-learning calibration model based on the updated historical mapping dataset using a transfer learning algorithm.

[0038] Example 7, see appendix Figure 1 This embodiment, based on the above embodiment, dynamically adjusts the target belt scale in the dynamic control module, specifically including: The second measurement feature is compared with the target set value to obtain the error value. The target setpoint is determined by the benchmark metrology curve generated by the self-learning calibration module; In one possible implementation, based on the error value Calculate the output control value Specifically, it is expressed as: in, This is expressed as a proportionality coefficient. Represented as integral coefficient, Represented as differential coefficients, Indicated as an index for the current time. This is represented as an index within the time range during integration calculation; It should be noted that the proportional coefficient, integral coefficient, and derivative coefficient are preset according to the actual operating state of the target belt scale; based on the output control value... Adjust the corresponding components of the target belt scale; if the belt running speed of the second metering characteristic is abnormal, then... This is converted into a speed regulation signal for the belt-driven motor; if the belt tension is abnormal, then... This is converted into a control signal for the belt tensioning device.

[0039] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A high-precision belt scale based on artificial intelligence that offers long-term stability and reliability, characterized in that: include: Environmental sensing module: used to monitor the operating environment data of the target belt scale in real time through multiple types of high-precision acquisition devices; Data acquisition module: Based on the high-precision acquisition device, acquire the first measurement data of the target belt scale; Multimodal data fusion module: used to perform deep fusion processing on the first measurement data and the operating environment data through a multimodal data fusion algorithm to obtain the first measurement feature corresponding to the first measurement data; Self-learning calibration module: used to acquire a self-learning calibration model, input the first metrological feature into the self-learning calibration model for self-learning and self-calibration, and acquire the second metrological data; Intelligent fault early warning module: used to perform real-time analysis and monitoring of the second metering data through pre-built fault early warning rules, and obtain the second metering characteristics of the target belt scale; Dynamic control module: Based on the second metering feature, dynamically adjust the target belt scale.

2. The high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 1, characterized in that: The environmental sensing module provides operating environment data for the target belt scale, including ambient temperature and humidity values ​​collected by a temperature and humidity sensor, acceleration, displacement, and velocity of mechanical vibration in various directions collected by a three-dimensional vibration sensor, dynamic dust concentration data collected by a laser dust sensor, and electromagnetic interference signals collected by an electromagnetic interference detection device.

3. The high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 1, characterized in that: The data acquisition module provides the target belt scale's first measurement data, which includes dynamic weight signals, speed fluctuation components, and belt tension.

4. The high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 1, characterized in that: The self-learning calibration module acquires a self-learning calibration model, specifically including: Construct a dataset containing historical mapping relationships between environmental parameters and corresponding measurement deviations; Historical spatiotemporal features of the historical mapping relationship dataset are extracted using a convolutional neural network. A transfer learning algorithm is used to iteratively train a deep neural network model using the historical spatiotemporal features to obtain a self-learning calibration model.

5. A high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 1, characterized in that: The self-learning calibration module acquires the second measurement data, specifically including: A benchmark measurement curve is generated based on the first measurement feature; Compensation parameters are generated by combining a self-learning calibration model; The compensation parameters are spatiotemporally aligned with the first measurement data, and calibration data is generated through a weighted fusion algorithm. After confidence verification, the calibration data is output.

6. A high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 5, characterized in that: The self-learning calibration module calculates the time based on the benchmark metrology curve corresponding to the first metrology feature. Compensation parameters at time Specifically, it is expressed as: in, Represented as in time The measurement deviation between the actual measured value and the reference measured value Represented as the baseline measurement curve, Represented as the current time index, This is represented by the preset environmental compensation coefficient. This is represented as the preset trend tracking coefficient.

7. A high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 1, characterized in that: The intelligent fault early warning module has pre-built fault early warning rules, specifically including: Data fluctuation threshold rules based on statistical process control; Wear and degradation prediction trend based on the cumulative operating time of the target belt scale; The correlation rules between temperature gradient anomalies and mechanical deformation; The data fluctuation threshold of statistical process control is the dynamic error data generated by comparing the theoretical value and the measured value of the weights, which are automatically loaded into the weighing area of ​​the target belt scale by a servo motor.

8. A high-precision belt scale based on artificial intelligence with long-term stability and reliability as described in claim 7, characterized in that: The intelligent fault early warning module, which uses a servo motor to automatically load weights, specifically includes: The preset permutation and combination modes include single-point loading, multi-point symmetrical loading, and dynamic moving loading. Based on the coupling relationship between the preset permutation and combination pattern and the belt speed, the theoretical weight distribution curve is calculated to dynamically correct the self-learning calibration model.

Images