SIP acquisition system of six-dimensional force sensor

Abstract

The invention relates to an SIP (Session Initiation Protocol) acquisition system of a six-dimensional force sensor, and the system comprises an ADC (Analog to Digital Converter) acquisition processing module which is used for acquiring six paths of analog signals of force and torque and converting the six paths of analog signals into six paths of digital signals; the multi-mode monitoring module is used for collecting a temperature signal, an electromagnetic interference signal and a vibration signal; the MCU processing module is used for performing digital filtering and calibration on the six paths of digital signals, the temperature signals, the electromagnetic interference signals and the vibration signals, decoupling the calibrated six paths of digital signals based on a trained decoupling model, and outputting a six-dimensional force decoupling value after temperature, vibration and electromagnetic interference are removed; and the ADC acquisition processing module, the multi-mode monitoring module and the MCU processing module are packaged by adopting an SIP (Session Initiation Protocol) technology. According to the invention, the SIP technology is adopted to package each module in the system, and the trained decoupling model is deployed in the MCU processing module, so that the problem that the size of the sensor acquisition system is large is effectively solved, and the accuracy of a measurement result is ensured.

Description

Technical Field

[0001] The present invention relates to the field of sensor technology, and in particular to a SIP acquisition system of a six-dimensional force sensor. Background Art

[0002] The six-axis force sensor can simultaneously measure forces in three directions (Fx, Fy, Fz) and moments in three directions (Mx, My, Mz), and plays a vital role in the fields of robotics, aerospace, biomedical engineering, precision manufacturing, etc. The acquisition system of a traditional six-axis force sensor is usually composed of multiple discrete components, including analog-to-digital converters and processors, which results in a relatively large volume and is difficult to meet scenarios with strict requirements on application space. At the same time, the coupling error of the six-axis force sensor will affect its detection accuracy, and the deviation of the decoupling result will also make it difficult to meet application scenarios with high requirements on measurement accuracy.

[0003] Therefore, providing a six-dimensional force sensor acquisition system to ensure accurate and efficient signal acquisition and processing in a limited space is an urgent problem to be solved. Summary of the invention

[0004] In view of this, the main purpose of the present invention is to provide a SIP acquisition system for a six-dimensional force sensor, which uses SIP technology to encapsulate the modules in the acquisition system and deploys a trained decoupling model in the MCU processing module, effectively solving the problem of the large size of the sensor acquisition system and ensuring the accuracy of the measurement results.

[0005] The embodiment of the present application provides a SIP acquisition system of a six-dimensional force sensor, including: An ADC acquisition processing module is used to acquire six analog signals of force and torque and convert the six analog signals into six digital signals; Multimodal monitoring module, used to collect temperature signals, electromagnetic interference signals and vibration signals; The MCU processing module is used to digitally filter and calibrate the six-channel digital signals, the temperature signal, the electromagnetic interference signal and the vibration signal, and decouple the calibrated six-channel digital signals based on the trained decoupling model, and output the six-dimensional force decoupling value after removing the temperature, vibration and electromagnetic interference; Among them, the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module are packaged using SIP technology.

[0006] In some embodiments, the ADC acquisition processing module includes: ADC data acquisition interface, used to collect six analog signals of force and torque; Anti-aliasing filter, used to filter out high-frequency noise from six-channel analog signals; A PGA amplifier is used to adjust the gain of six analog signals after filtering out high-frequency noise; The ADC chip is used for converting the six-channel analog signals after gain adjustment into six-channel digital signals.

[0007] In some embodiments, the multimodal monitoring module includes: A temperature monitoring module is used to collect temperature signals and send the temperature signals to the MCU processing module; An electromagnetic monitoring module, used for collecting electromagnetic interference signals and sending the electromagnetic interference signals to the MCU processing module; The vibration monitoring module is used to collect vibration signals and send the vibration signals to the MCU processing module.

[0008] In some embodiments, it also includes: A data cache module, used for temporarily storing the six digital signals, the temperature signal, the electromagnetic interference signal and the vibration signal for reading by the MCU processing module; A crystal oscillator circuit, used to provide a clock signal for the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module; The RS485 data communication interface is used to output the six-dimensional force decoupling value obtained by the MCU processing module.

[0009] In some of the embodiments, when the SIP acquisition system is packaged using SIP technology, the substrate material used is aluminum nitride ceramic, and the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module are stacked and packaged in three dimensions based on TSV interconnection technology.

[0010] In some embodiments, the training step of the decoupling model includes: Collecting sensor output data and true force value data corresponding to different working conditions, as well as corresponding temperature data, electromagnetic interference data and vibration data to obtain a first data set; After normalizing the first data set, a second data set is obtained; Dynamically assign corresponding weights to the sensor data, temperature data, electromagnetic interference data, and vibration data in the second data set, and obtain a fused feature vector after weighted fusion; The fused feature vector is input into the neural network, the error is calculated according to the loss function, and the network parameters are updated by using the optimization algorithm. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module.

[0011] In some embodiments, the first data set is normalized sequentially to obtain the second data set, including: Normalizing each data in the first data set respectively, mapping the data to the interval [0, 1] or [-1, 1], to obtain a second data set; Among them, the normalization formula is:

[0012] or ; In the formula Represents the normalized data, Indicates the minimum value in the data. Indicates the maximum value in the data.

[0013] In some embodiments, dynamically assigning corresponding weights to the sensor data, temperature data, electromagnetic interference data, and vibration data in the second data set, and obtaining a fused feature vector after weighted fusion, includes: Perform polynomial fitting on the normalized data to calculate the corresponding weight adjustment coefficient; and the weight adjustment coefficient includes the temperature electromagnetic coupling coefficient α(T), the temperature vibration coupling coefficient β(T), the electromagnetic vibration coupling coefficient γ ( E ) and vibration electromagnetic coupling coefficient δ(V); According to different working conditions and the corresponding weight adjustment coefficient, the weight of each feature is dynamically adjusted to obtain the temperature data weight, vibration data weight and electromagnetic interference weight; among them, Temperature data weight ; In the formula is the benchmark weight of temperature data; Vibration data weight ; In the formula is the baseline weight of the vibration data; Electromagnetic interference data weight ; In the formula is the baseline weight of electromagnetic interference data; After weighted fusion of the temperature data weight, the vibration data weight and the electromagnetic interference weight, a fused feature vector is obtained; and the fused feature vector is: ; In the formula The base weight of the sensor output data, is the normalized sensor data, is the normalized temperature data, is the normalized electromagnetic interference data, is the normalized vibration data.

[0014] In some embodiments, the fused feature vector is input into the neural network, the error is calculated according to the loss function, and the network parameters are updated by the optimization algorithm. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module, including: Inputting the fused feature vector into a neural network to calculate a predicted value; The error between the predicted value and the normalized actual force data is calculated using the mean square error, and the weights and biases of the neural network are updated using a target optimization algorithm to minimize the loss function; Among them, the formula of the loss function is: ; In the formula is the true value, is the predicted value, is the L2 regularization coefficient, is the number of training samples, is the neural network weight, is the total number of model weight parameters; Iterate the training until the loss function converges or reaches a preset number of training rounds, and obtain a trained neural network model, which is deployed in the MCU processing module.

[0015] In some embodiments, the target optimization algorithm is any one of SGD, Adagrad or Adam.

[0016] Technical effects of the present invention: The present application provides a SIP acquisition system for a six-dimensional force sensor, including an ADC acquisition processing module for collecting six analog signals of force and torque, and converting the six analog signals into six digital signals; a multimodal monitoring module for collecting temperature signals, electromagnetic interference signals and vibration signals; an MCU processing module for digitally filtering and calibrating the six digital signals, temperature signals, electromagnetic interference signals and vibration signals, and decoupling the calibrated six digital signals based on a trained decoupling model, and outputting six-dimensional force decoupling values ​​after removing temperature, vibration and electromagnetic interference; wherein the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module are packaged using SIP technology. The present application uses SIP technology to encapsulate the modules in the acquisition system, which can effectively improve the large volume problem caused by the traditional acquisition system composed of discrete components, and deploys a trained decoupling model in the MCU processing module, comprehensively considering the influence of temperature, vibration and electromagnetic interference, ensuring the accuracy of the measurement results, and further meeting the high-precision measurement requirements.

[0017] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0019] Figure 1 A schematic block diagram of the structure of a SIP acquisition system of a six-dimensional force sensor according to an embodiment of the present application is shown; Figure 2 Shown is a flowchart of the training steps of the decoupled model according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0021] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0022] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0023] As a key device that can measure forces and torques in three-dimensional space, six-dimensional force sensors play an important role in many fields such as industrial robots, aerospace, medical rehabilitation, etc., and in these application scenarios, accurately obtaining force and torque information is crucial for the reliable operation of the system. It should be noted that six-dimensional force sensors usually measure forces and torques based on the deformation of elastic bodies in actual work. When forces and torques in various directions act on the elastic body, the elastic body will be deformed, resulting in coupling effects between different measurement channels, affecting the measurement accuracy. Therefore, it is necessary to parse independent six-dimensional force and torque components from the original signal to ensure the accuracy of the measurement results. However, in the actual decoupling process, the decoupling result is susceptible to a variety of interference factors, especially inside the sensor, the internal signal interference caused by the complex circuit structure directly acts on the generation and transmission process of the measurement signal, which has a direct impact on the decoupling result. Therefore, this application specifically designs a SIP acquisition system for a six-dimensional force sensor for the above-mentioned problems to eliminate or reduce the influence of internal interference signals on the decoupling results, and further meet the high-precision measurement requirements of the six-dimensional force sensor.

[0024] like Figure 1 As shown, the embodiment of the present application provides a SIP acquisition system of a six-dimensional force sensor, including: The ADC acquisition processing module 100 is used to acquire six analog signals of force and torque and convert the six analog signals into six digital signals; The multi-modal monitoring module 200 is used to collect temperature signals, electromagnetic interference signals and vibration signals; The MCU processing module 300 is used to digitally filter and calibrate the six-channel digital signals, the temperature signal, the electromagnetic interference signal and the vibration signal, and decouple the calibrated six-channel digital signals based on the trained decoupling model, and output the six-dimensional force decoupling value after removing the temperature, vibration and electromagnetic interference; Among them, the ADC acquisition processing module 100, the multimodal monitoring module 200 and the MCU processing module 300 are packaged using SIP technology.

[0025] This embodiment provides a SIP acquisition system for a six-dimensional force sensor, which uses SIP technology to integrate the ADC processing acquisition module 100, the MCU processing module 300 and the multimodal monitoring module 200 in one package. Compared with the existing acquisition system composed of multiple discrete components, the volume can be reduced by several times or dozens of times, and it is more suitable for use in scenarios with strict space requirements. Specifically, the ADC acquisition processing module 100 is responsible for collecting the differential output of 6 groups of strain gauges and converting the corresponding 6 analog signals into 6 digital signals, and the MCU processing module 300 is responsible for processing the 6 digital signals.

[0026] It should be noted that a trained decoupling model is deployed in the MCU processing module 300 of this embodiment. In the decoupling process of six-dimensional force and torque, the trained decoupling model is used to comprehensively consider the influence of temperature signal, electromagnetic interference signal and vibration signal on the decoupling result, and a more accurate six-dimensional force decoupling value is calculated. Among them, after the collected data are input into the MCU processing module 300, data filtering, calibration and decoupling processing are performed in sequence. Specifically, the digital filtering uses an FIR filter to remove part of the noise interference in the signal, and calibrates and eliminates the zero-point error of the sensor to improve the accuracy of the measurement. Finally, the trained decoupling model is used to eliminate the coupling effect between the channels of the six-dimensional force sensor to obtain more accurate force and torque information. It should be noted here that since the FIR filter can only remove noise of a specific frequency in the signal, the interference removal related to low-frequency temperature drift and low-frequency components of electromagnetic interference is limited, and the influence of nonlinear temperature, electromagnetic interference and vibration on the force signal cannot be processed, and the signal calibration has limited effect on dynamic interference. Therefore, this embodiment specially designs a decoupling model to comprehensively consider the influence of temperature signals, electromagnetic interference signals and vibration signals on the force signal, handle nonlinear relationships, and can dynamically adjust according to real-time collected data to ensure the accuracy of the decoupling results to meet high-precision measurement requirements.

[0027] That is, the SIP acquisition system of the six-axis force sensor of this embodiment not only adopts the SIP integrated packaging method in the hardware setting, which makes the whole system smaller, but also uses the specially trained decoupling model deployed by the MCU processing module 300 to eliminate the coupling influence between the channels of the six-axis force sensor. In this process, the influence of temperature, vibration, and electromagnetic interference generated by the internal circuit of the sensor on the decoupling result is comprehensively considered, so that the processing result of the acquisition system is more accurate and the system reliability is improved.

[0028] In some embodiments, the ADC acquisition processing module 100 includes: ADC data acquisition interface, used to collect six analog signals of force and torque; Anti-aliasing filter, used to filter out high-frequency noise from six-channel analog signals; A PGA amplifier is used to adjust the gain of six analog signals after filtering out high-frequency noise; The ADC chip is used for converting the six-channel analog signals after gain adjustment into six-channel digital signals.

[0029] In this embodiment, the ADC acquisition processing module 100 is used to convert the collected analog signal into a digital signal. Among them, the ADC acquisition interface in the ADC acquisition processing module 100 collects 6 analog signals, and transmits the 6 analog signals to the anti-aliasing filter for preliminary processing to eliminate the high-frequency noise therein. The output end of the anti-aliasing filter is connected to the input end of the PGA amplifier, so that the PGA amplifier performs gain adjustment on the 6 analog signals passing through the anti-aliasing filter to meet the requirements of the ADC chip for the input signal amplitude. The PGA amplifier inputs the gain-adjusted signal to the ADC chip to convert it into 6 digital signals for subsequent decoupling. It should be noted that the ADC acquisition processing module 100 of this embodiment preliminarily eliminates high-frequency noise interference and prevents aliasing effects through reasonable anti-aliasing filtering and gain adjustment, and after gain adjustment, the signal amplitude is within the optimal input range of the ADC chip to improve the signal resolution. The ADC chip converts the analog signal into a digital signal, which is convenient for subsequent decoupling algorithm processing and data analysis. Compared with analog signals, digital signals have stronger anti-interference ability and are easy to transmit, which can effectively improve processing efficiency.

[0030] It should be noted that the PGA amplifier and the ADC chip of this embodiment are arranged close to each other so that the analog signal output by the PGA can be transmitted to the ADC chip as quickly as possible for analog-to-digital conversion, shortening the distance between the two to reduce the attenuation during signal transmission. At the same time, avoid placing the MCU chip too close to the temperature-sensitive PGA amplifier or ADC chip to affect the performance of the PGA amplifier or ADC chip. In addition, each part of the ADC acquisition processing module 100 is packaged in SIP.

[0031] It should also be noted that in this embodiment, differential signal wiring is used between the ADC acquisition processing module 100 and the MCU processing module 300, and the lengths of the differential signal lines are equal and the spacing is uniform to ensure the balance of the differential signal.

[0032] In some embodiments, the multimodal monitoring module 200 includes: A temperature monitoring module is used to collect temperature signals and send the temperature signals to the MCU processing module 300; The electromagnetic monitoring module is used to collect electromagnetic interference signals and send the electromagnetic interference signals to the MCU processing module 300; The vibration monitoring module is used to collect vibration signals and send the vibration signals to the MCU processing module 300 .

[0033] In this embodiment, the temperature signal, electromagnetic interference signal and vibration signal in the monitoring system are monitored by setting a multimodal monitoring module 200 to reduce the influence of internal interference factors generated by the circuit in the system during operation on the decoupling result. Specifically, when the sensor is working, its internal electronic components, such as ADC chip, MCU processing module 300, etc., will generate heat accordingly, resulting in an increase in internal temperature, especially the MCU processing module 300, which has relatively large power consumption and relatively more heat generated accordingly, and introduces additional errors to affect the decoupling result. In addition, the digital signal will also generate certain electromagnetic interference during the transmission process, resulting in inaccurate decoupling results. At the same time, the deformation of the strain gauge when sensing the external force will inevitably generate vibration signals. Therefore, in this embodiment, a temperature monitoring module is set to collect temperature signals, an electromagnetic monitoring module is set to collect electromagnetic interference signals, and a vibration monitoring module is set to collect vibration signals, and a trained decoupling model is used to remove the influence of temperature, vibration, and electromagnetic interference on the force signal, thereby restoring the true force value.

[0034] In some embodiments, it also includes: A data cache module, used for temporarily storing six digital signals, temperature signals, electromagnetic interference signals and vibration signals for reading by the MCU processing module 300; A crystal oscillator circuit, used to provide a clock signal for the ADC acquisition processing module 100, the multi-modal monitoring module 200 and the MCU processing module 300; The RS485 data communication interface is used to output the six-dimensional force decoupling value processed by the MCU processing module 300.

[0035] In this embodiment, a data cache module, a crystal oscillator circuit and an RS485 data communication interface are also provided, wherein a stable clock signal is provided to the system through the crystal oscillator circuit to ensure that each module runs synchronously. The digital cache module is used to temporarily store the digital signal output by the ADC acquisition processing module 100 so that the MCU processing module 300 can read it on demand, and can also store the temperature, electromagnetic and vibration signals collected by the multimodal monitoring module 200. It should also be noted that this embodiment converts the six-dimensional force decoupling value output by the MCU processing module 300 into an RS485 signal by setting an RS485 data communication interface, and transmits it to the outside through the RS485 bus, thereby realizing the integration of acquisition, storage and calculation.

[0036] In some of the embodiments, when the SIP acquisition system is packaged using SIP technology, the substrate material used is aluminum nitride ceramic, and the ADC acquisition processing module 100, the multimodal monitoring module 200 and the MCU processing module 300 are stacked and packaged in three dimensions based on TSV interconnection technology.

[0037] The SIP acquisition system of this embodiment uses TSV technology to stack and package the ADC acquisition processing module 100, the multimodal monitoring module 200 and the MCU processing module 300, so as to achieve chip stacking in three dimensions, which can provide a short-distance, high-speed signal transmission channel, reduce the package volume, and improve system performance. At the same time, since aluminum nitride ceramics have high mechanical strength, good insulation performance and higher thermal conductivity, aluminum nitride ceramics are used as the substrate to dissipate the heat generated by the system and alleviate the influence of heat on the decoupling results. At the same time, the thermal expansion coefficient of the ceramic substrate material is close to that of the chip, which can also reduce the problem of interconnection failure caused by thermal stress.

[0038] like Figure 2 As shown, in some embodiments, the training steps of the decoupled model include: S100, collecting sensor output data and real force value data corresponding to different working conditions, as well as corresponding temperature data, electromagnetic interference data and vibration data, to obtain a first data set; S200, normalizing the first data set to obtain a second data set; S300, dynamically assigning corresponding weights to the sensor data, temperature data, electromagnetic interference data, and vibration data in the second data set, and obtaining a fused feature vector after weighted fusion; S400, input the fused feature vector into the neural network, calculate the error according to the loss function, and use the optimization algorithm to update the network parameters. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module 300.

[0039] In this embodiment, the original output data under different working conditions are collected, and various force and torque combinations should be covered as much as possible during the collection process to ensure the comprehensiveness and representativeness of the data. Specifically, different forces and torques can be applied in different directions, and the specific values ​​of the applied forces and torques, as well as the output values ​​of the corresponding sensors, are recorded. At the same time, in order to eliminate the influence of temperature, vibration and electromagnetic interference in the decoupling process, the corresponding temperature data, electromagnetic interference data and vibration data are also collected to construct a first data set. Further, the constructed first data set is normalized, and the data is mapped to the interval [0, 1] or [-1, 1] to obtain a second data set to speed up the training speed. Furthermore, the corresponding weights are dynamically assigned to the temperature data, electromagnetic interference data, vibration data and sensor data in the second data set, and weighted fusion is performed to reflect the influence of temperature, vibration and electromagnetic interference on the sensor output. Finally, the obtained fusion feature vector is input into the neural network, the loss error is calculated, and the optimization algorithm is used to optimize and update the network parameters. The above training process is repeated until the loss function converges or the preset number of training times is reached, thereby obtaining a trained decoupling model, which is deployed in the MCU processing module 300 to achieve real-time decoupling of six-dimensional forces and torques.

[0040] The training process of the decoupled model is described in detail below.

[0041] In some embodiments, the first data set is normalized in sequence to obtain the second data set, including: Each data in the first data set is normalized and mapped to the interval [0, 1] or [-1, 1] to obtain a second data set; Among them, the normalization formula is:

[0042] or ; In the formula Represents the normalized data, Represents the original data, Indicates the minimum value in the data. Indicates the maximum value in the data.

[0043] In this embodiment, the collected data is normalized so that the data is in a relatively uniform scale range to speed up the model convergence speed. Specifically, after the 6-channel analog signal collected by the ADC acquisition processing module 100 is filtered, amplified and converted into 6-channel digital signals, it is input into the MCU processing module 300 for digital filtering and calibration. At the same time, the temperature data, vibration data and electromagnetic interference data collected by the multimodal monitoring module 200 are also input into the MCU processing module 300 for digital filtering and calibration, and the first data set is constructed with the calibrated data. Subsequently, each data in the first data set is normalized, and the data is mapped to the [0, 1] interval or the [-1, 1] interval. In addition, the normalized data is used as the second data set for subsequent processing of the fusion feature vector. That is, when defining the weight coupling coefficient later, the normalized data is used to eliminate the influence of the data dimension and make the coupling relationship between different features clearer.

[0044] In some embodiments, corresponding weights are dynamically assigned to the sensor data, temperature data, electromagnetic interference data, and vibration data in the second data set, and a fused feature vector is obtained after weighted fusion, including: The normalized data are fitted with a polynomial to calculate the corresponding weight adjustment coefficients; the weight adjustment coefficients include the temperature electromagnetic coupling coefficient α(T), the temperature vibration coupling coefficient β(T), and the electromagnetic vibration coupling coefficient γ ( E ) and vibration electromagnetic coupling coefficient δ(V); According to different working conditions and the corresponding weight adjustment coefficient, the weight of each feature is dynamically adjusted to obtain the temperature data weight, vibration data weight and electromagnetic interference weight; among them, Temperature data weight ; In the formula is the benchmark weight of temperature data; Vibration data weight ; In the formula is the baseline weight of the vibration data; Electromagnetic interference data weight ; In the formula is the baseline weight of electromagnetic interference data; After weighted fusion of temperature data weight, vibration data weight and electromagnetic interference weight, a fused feature vector is obtained; and the fused feature vector is: ; In the formula The base weight of the sensor output data, is the normalized sensor data, is the normalized temperature data, is the normalized electromagnetic interference data, is the normalized vibration data.

[0045] In this embodiment, feature extraction is performed before weighted fusion, and weighted fusion processing is performed on the extracted features. Among them, in order to consider the mutual influence between temperature, electromagnetic interference and vibration, a dynamic weight allocation method is used to adjust the weight of each feature according to different working conditions to improve the adaptability of the algorithm. Under different working conditions, different forces and torques are applied, and the power consumption of the components inside the sensor is different. As the power consumption increases, more heat will be generated accordingly, bringing about temperature changes. In particular, the SIP acquisition system of this embodiment uses SIP technology to package and integrate each module together, and there is a possibility of local heat concentration, which is more obvious under high load or long-term operation. In addition, the electronic components inside the sensor will generate an electromagnetic field when working. As the applied force and torque change, the corresponding electromagnetic interference signal will also change accordingly, and the frequency of digital signal transmission will change, bringing different degrees of electromagnetic interference, and the obvious degree of vibration will also vary with the size of the applied force. Moreover, under different working conditions, the magnitude of the applied force and torque is different. Temperature, electromagnetic interference and vibration are not independent and may affect each other. Temperature changes may cause changes in circuit component parameters, thereby affecting electromagnetic interference characteristics or vibration characteristics. There is also mutual influence between electromagnetic interference and vibration interference. A simple linear relationship cannot accurately describe the relationship between them. Based on this, this embodiment sets the temperature electromagnetic coupling coefficient , temperature vibration coupling coefficient β(T), electromagnetic vibration coupling coefficient γ ( E ) and the vibration electromagnetic coupling coefficient δ(V) are used to quantify the mutual influence between temperature, electromagnetic interference and vibration under different working conditions, so as to more accurately reflect their influence on force measurement. In addition, the weight of each feature is dynamically adjusted according to the weight adjustment coefficient to improve the accuracy and robustness of the decoupling model. Finally, the features after dynamic weight adjustment are weighted fused to obtain a fused feature vector, which is a feature vector that can fully reflect the current working conditions, so as to more accurately remove the corresponding interference and further meet the needs of high-precision measurement.

[0046] Specifically, the coupling relationship model is established by measuring the sensor output data under different conditions, and the polynomial fitting data is used to obtain the data expression of the weight adjustment coefficient, the temperature-electromagnetic coupling coefficient , temperature-vibration coupling coefficient , electromagnetic-vibration coupling coefficient , vibration-electromagnetic coupling coefficient , where a, b, c, d, e, f, j, k, l are all fitting coefficients, which can be solved by the least square method. Further, after obtaining the weight adjustment coefficient, each feature weight is dynamically adjusted according to the temperature data, electromagnetic interference data and vibration data collected under different working conditions to reflect their influence on the sensor output, and the temperature data weight, vibration data weight and electromagnetic interference data weight are obtained respectively. Finally, the sensor data, temperature, electromagnetic interference and vibration data are fused by the weighted fusion formula to form a comprehensive vector to ensure that the decoupling model can take these interference factors into account, so that in actual applications, after collecting force and torque data and corresponding temperature data, electromagnetic interference data and vibration data, the trained decoupling model can be input to obtain the real six-dimensional force decoupling value. That is, in the feature fusion process, this embodiment dynamically adjusts the weights of each feature, so that the influence of temperature, electromagnetic interference and vibration factors on the sensor output can be more accurately reflected, thereby improving the accuracy and robustness of the decoupling algorithm.

[0047] In some embodiments, the fused feature vector is input into the neural network, the error is calculated according to the loss function, and the network parameters are updated by the optimization algorithm. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module 300, including: Input the fused feature vector into the neural network to calculate the predicted value; The error between the predicted value and the normalized actual force data is calculated using the mean square error, and the weights and biases of the neural network are updated using a target optimization algorithm to minimize the loss function; Among them, the formula of the loss function is: ; In the formula is the true value, is the predicted value, is the L2 regularization coefficient, is the number of training samples, is the neural network weight, is the total number of model weight parameters; Iterate the training until the loss function converges or reaches a preset number of training rounds, and obtain a trained neural network model, which is deployed in the MCU processing module 300.

[0048] In this embodiment, after obtaining the fused feature vector, it is input into the neural network for training to learn the nonlinear relationship between input and output, wherein the error between the predicted value and the true value is calculated by combining the mean square error with L2 regularization, and the gradient of the loss function to the model parameters is calculated by back propagation, and the weight and bias of the neural network are updated. The addition of L2 regularization can prevent the model from overfitting, and improve the numerical stability of the training, so that the model can learn a smoother weight distribution, so as to better remove the influence of temperature, electromagnetic interference and vibration on the decoupling results. During the training process, the above training process is repeated many times until the loss function converges or reaches the preset number of training times, and a trained neural network model is obtained, which is deployed in the MCU processing module 300 of the SIP acquisition system of the six-dimensional force sensor, so that the six-channel real-time data is input into the neural network, and the real six-dimensional force and torque after decoupling can be obtained through the network calculation.

[0049] In some of the embodiments, the target optimization algorithm is any one of SGD, Adagrad or Adam.

[0050] In this embodiment, the optimization algorithm is selected as any one of the SGD stochastic gradient descent optimization algorithm, the Adagrad adaptive gradient algorithm or the Adam adaptive learning rate optimization algorithm, wherein the SGD stochastic gradient descent optimization algorithm is more preferred, which has adaptive learning ability, can automatically adjust the learning rate during the training process, accelerate the convergence speed, and is suitable for large-scale data sets and complex nonlinear models. Through the optimization of SGD, the decoupling model of this embodiment can gradually adjust the parameters, remove the influence of temperature, electromagnetic interference and vibration on the force measurement results, and improve the measurement accuracy of the sensor.

[0051] In summary, in high-precision measurement applications, such as industrial robots, precision machining, etc., even small temperature changes, electromagnetic interference or vibration interference can cause measurement errors, and in practical applications, temperature, electromagnetic interference and vibration often exist at the same time, and there may be a complex coupling relationship between the three. Therefore, the present application comprehensively considers different interference factors, trains a decoupling model, and deploys it in the MCU processing module 300 to improve the measurement accuracy of the sensor under different working conditions and meet the use requirements. At the same time, each module in the acquisition system of the present application is packaged using SIP technology, which effectively reduces the overall size of the acquisition system. The miniaturized design is not only easy to install and use, but also can effectively broaden the scope of application and be integrated into more compact devices. The encapsulated design can also better prevent the external environment from affecting it. In addition, the modular design is easier to maintain and upgrade.

[0052] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0053] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "one", "the" and the like involved in this application do not indicate a quantitative limitation and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more, and "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. The terms "first", "second", "third" and the like involved in this application are merely to distinguish similar objects and do not represent a specific ordering of objects.

[0054] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but the limitation of the scope of the invention patent cannot be understood as a result. It should be pointed out that, for a person of ordinary skill in the art, several variations or improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A SIP acquisition system for a six-dimensional force sensor, characterized in that: include: An ADC acquisition processing module is used to acquire six analog signals of force and torque and convert the six analog signals into six digital signals; Multimodal monitoring module, used to collect temperature signals, electromagnetic interference signals and vibration signals; The MCU processing module is used to digitally filter and calibrate the six-channel digital signals, the temperature signal, the electromagnetic interference signal and the vibration signal, and decouple the calibrated six-channel digital signals based on the trained decoupling model, and output the six-dimensional force decoupling value after removing the temperature, vibration and electromagnetic interference; Among them, the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module are packaged using SIP technology.

2. The SIP acquisition system of the six-dimensional force sensor according to claim 1 is characterized in that: The ADC acquisition processing module includes: ADC data acquisition interface, used to collect six analog signals of force and torque; Anti-aliasing filter, used to filter out high-frequency noise from six-channel analog signals; A PGA amplifier is used to adjust the gain of six analog signals after filtering out high-frequency noise; The ADC chip is used for converting the six-channel analog signals after gain adjustment into six-channel digital signals.

3. The SIP acquisition system of the six-dimensional force sensor according to claim 1 is characterized in that: The multimodal monitoring module comprises: A temperature monitoring module is used to collect temperature signals and send the temperature signals to the MCU processing module; An electromagnetic monitoring module, used for collecting electromagnetic interference signals and sending the electromagnetic interference signals to the MCU processing module; The vibration monitoring module is used to collect vibration signals and send the vibration signals to the MCU processing module.

4. The SIP acquisition system of the six-dimensional force sensor according to claim 1, characterized in that: Also includes: A data cache module, used for temporarily storing the six digital signals, the temperature signal, the electromagnetic interference signal and the vibration signal for reading by the MCU processing module; A crystal oscillator circuit, used to provide a clock signal for the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module; The RS485 data communication interface is used to output the six-dimensional force decoupling value obtained by the MCU processing module.

5. The SIP acquisition system of the six-dimensional force sensor according to claim 1, characterized in that: When the SIP acquisition system is packaged using the SIP technology, the substrate material used is aluminum nitride ceramic, and the ADC acquisition processing module, the multimodal monitoring module and the MCU processing module are stacked and packaged in three dimensions based on the TSV interconnection technology.

6. The SIP acquisition system of the six-dimensional force sensor according to any one of claims 1 to 5, characterized in that: The training steps of the decoupled model include: Collecting sensor output data and true force value data corresponding to different working conditions, as well as corresponding temperature data, electromagnetic interference data and vibration data to obtain a first data set; After normalizing the first data set, a second data set is obtained; Dynamically assign corresponding weights to the sensor data, temperature data, electromagnetic interference data, and vibration data in the second data set, and obtain a fused feature vector after weighted fusion; The fused feature vector is input into the neural network, the error is calculated according to the loss function, and the network parameters are updated by using the optimization algorithm. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module.

7. The SIP acquisition system of the six-dimensional force sensor according to claim 6, characterized in that: The first data set is normalized in sequence to obtain a second data set, including: Normalizing each data in the first data set, mapping the data to the interval [0, 1] or [-1, 1], to obtain a second data set; Among them, the normalization formula is: or In the formula Represents the normalized data, Indicates the minimum value in the data. Indicates the maximum value in the data.

8. The SIP acquisition system of the six-dimensional force sensor according to claim 7, characterized in that: The dynamically assigning corresponding weights to the sensor data, temperature data, electromagnetic interference data and vibration data in the second data set, and obtaining a fused feature vector after weighted fusion, includes: Perform polynomial fitting on the normalized data to calculate the corresponding weight adjustment coefficient; and the weight adjustment coefficient includes the temperature electromagnetic coupling coefficient α(T), the temperature vibration coupling coefficient β(T), the electromagnetic vibration coupling coefficient γ ( E ) and vibration electromagnetic coupling coefficient δ(V); According to different working conditions and the corresponding weight adjustment coefficient, the weight of each feature is dynamically adjusted to obtain the temperature data weight, vibration data weight and electromagnetic interference weight; among them, Temperature data weight ; In the formula is the benchmark weight of temperature data; Vibration data weight ; In the formula is the base weight of vibration data; Electromagnetic interference data weight ; In the formula is the baseline weight of electromagnetic interference data; After weighted fusion of the temperature data weight, the vibration data weight and the electromagnetic interference weight, a fused feature vector is obtained; and the fused feature vector is: ； In the formula The base weight of the sensor output data, is the normalized sensor data, is the normalized temperature data, is the normalized electromagnetic interference data, is the normalized vibration data.

9. The SIP acquisition system of the six-dimensional force sensor according to claim 6, characterized in that: The fused feature vector is input into the neural network, the error is calculated according to the loss function, and the network parameters are updated by the optimization algorithm. After iterative training, a trained neural network model is obtained and deployed in the MCU processing module, including: Inputting the fused feature vector into a neural network to calculate a predicted value; The error between the predicted value and the normalized actual force data is calculated using the mean square error, and the weights and biases of the neural network are updated using a target optimization algorithm to minimize the loss function; Among them, the formula of the loss function is: ; In the formula is the true value, is the predicted value, is the L2 regularization coefficient, is the number of training samples, is the neural network weight, is the total number of model weight parameters; Iterate the training until the loss function converges or reaches a preset number of training rounds, and obtain a trained neural network model, which is deployed in the MCU processing module.

10. The SIP acquisition system of the six-dimensional force sensor according to claim 9, characterized in that: The target optimization algorithm is any one of SGD, Adagrad or Adam.

Images