Intelligent sensor self-diagnosis optimization system and intelligent sensor

Abstract

The intelligent sensor self-diagnosis optimization system comprises a laser interference unit and a temperature detection unit which are integrated in the sensor, the displacement phase difference of the sensor is obtained through laser interference measurement, and internal temperature data are combined to obtain the self-diagnosis optimization of the intelligent sensor. And the physical displacement is calculated in real time by a built-in artificial intelligence processor of the edge calculation unit. And the dynamic reference calibration unit is used for identifying reference workpiece characteristics in the production line based on the target detection neural network and generating a space error compensation matrix in combination with the physical displacement. And a compensation result is dynamically loaded by an FPGA register in the sensor compensation circuit unit, so that fine compensation of output signals of the sensor is realized. The method has high-precision dynamic error compensation, autonomous diagnosis optimization capability and good system integration, and is suitable for sensor adaptive control in scenes of high-precision manufacturing, intelligent factories and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial measurement and control and intelligent manufacturing, and in particular to an intelligent sensor self-diagnosis optimization system and an intelligent sensor. Background Art

[0002] With the continuous improvement of industrial automation, various smart sensors are playing a key role in manufacturing processes, equipment control, and quality monitoring. However, traditional smart sensors rely on static calibration methods, which makes it difficult to meet the requirements of high precision, high dynamic response, and adaptability to complex environments.

[0003] Currently, common sensor self-diagnosis technologies mainly rely on the following methods:

[0004] Electrical signal self-check mechanism: By regularly judging the output signal amplitude, voltage deviation, etc., a rough assessment of the sensor status is achieved, but the accuracy is limited;

[0005] Temperature compensation circuit: Some sensors embed simple thermal compensation modules to perform static calibration for ambient temperature drift, which cannot adapt to changes in multi-source coupling;

[0006] Manual regular calibration: Professionals are required to manually calibrate the error source with the help of a reference workpiece. This is not only inefficient but also easily affected by human factors.

[0007] Software filtering algorithms, such as Kalman filtering and sliding average, are used to suppress abnormal signals but cannot analyze the root cause of errors from a physical perspective.

[0008] In addition, as production line flexible manufacturing and dynamic working conditions (such as vibration and high temperature changes) continue to increase the requirements for positioning accuracy, traditional methods generally face the following limitations:

[0009] Lack of micro-nano-level precision acquisition capability: Ordinary capacitive and strain sensors have poor displacement measurement accuracy below the sub-micron level;

[0010] Unable to dynamically track error drift: Current compensation mechanisms are mostly "static calibration + periodic correction", which cannot achieve real-time dynamic compensation;

[0011] Lack of autonomous diagnosis and optimization mechanisms: Existing sensor systems are unable to automatically adjust internal compensation logic based on changes in the sensing environment and workpiece.

[0012] In recent years, some studies have attempted to introduce a combination of laser interferometry and artificial intelligence algorithms in the hope of achieving high-precision displacement diagnosis and intelligent compensation. However, there are still problems such as low system integration, insufficient real-time computing capabilities, and failure to form an integrated edge computing solution. Summary of the Invention

[0013] In view of the deficiencies in the prior art, the object of the present invention is to provide an intelligent sensor self-diagnosis optimization system and an intelligent sensor, which are used to achieve high-precision dynamic error compensation and autonomous diagnosis optimization of the intelligent sensor.

[0014] To achieve the above objectives, the present invention provides the following technical solutions: an intelligent sensor self-diagnosis optimization system, comprising:

[0015] A laser interferometer unit, integrated into the smart sensor, comprising a laser emitter emitting visible red light and a photon detector array, for acquiring the displacement phase difference of the smart sensor;

[0016] A temperature detection unit, integrated into the smart sensor, for detecting the internal temperature of the smart sensor;

[0017] an edge computing unit, communicatively connected to the laser interference unit and the temperature detection unit, and having an integrated artificial intelligence processor chip, configured to calculate a physical displacement of the smart sensor based on the displacement phase difference and the internal temperature;

[0018] a dynamic reference calibration unit, communicatively connected to the edge computing unit, configured to utilize an object detection neural network to identify preset reference workpiece features within the field of view of the smart sensor on the production line, and to generate a compensation matrix for spatial error compensation based on the preset reference workpiece features and the physical displacement;

[0019] The sensor compensation circuit unit includes an FPGA programmable register group, is communicatively connected to the dynamic reference calibration unit, and is used to dynamically configure the value of the FPGA programmable register group according to the compensation matrix to compensate the output signal of the smart sensor.

[0020] Furthermore, the photon detector array is a radially symmetric hexagonal close-packed structure, wherein a variable gain differential amplifier is provided between the central detector and the edge detector;

[0021] The edge computing unit includes:

[0022] an adjusting subunit, configured to dynamically adjust a gain coefficient of the variable gain differential amplifier based on the internal temperature data;

[0023] an operation subunit, configured to perform a temperature gradient-based phase unwrapping operation on the displacement phase difference;

[0024] The correction subunit is used to correct the displacement phase difference to eliminate thermo-optical distortion.

[0025] Furthermore, the calculation formula of the physical displacement is configured as:

[0026]

[0027] Wherein, represents the physical displacement, λ represents the laser wavelength of the laser transmitter, and Δφ represents

[0028] The displacement phase difference, α represents the phase compensation coefficient, erf represents the error function, β represents the temperature corresponding slope, T represents the internal temperature, T0 represents the preset reference temperature, γ represents the thermal expansion suppression factor, a represents the preset linear temperature drift coefficient, and b represents the preset nonlinear enhancement factor.

[0029] Furthermore, in the dynamic reference calibration unit, the target detection neural network is a dual-stream fusion architecture, and the dual-stream fusion architecture includes:

[0030] The first branch processes the visible light image features of the production line workpieces;

[0031] The second branch processes the depth phase point cloud generated by the laser interferometer unit;

[0032] The output of the dual-stream branch is passed through an attention-guided feature fusion module to generate the spatial coordinates of the preset reference workpiece features.

[0033] Furthermore, the dynamic reference calibration unit includes:

[0034] a decomposition subunit, configured to decompose the physical displacement into a rigid displacement component and a deformation displacement component;

[0035] an error calculation subunit, which calculates a speed-acceleration coupling error of a production line conveyor belt based on the preset reference workpiece feature;

[0036] A mapping subunit is connected to the error calculation subunit and the decomposition subunit, and is used to map the coupling error to the deformation displacement component through the Lie group SE (3) transformation model, and then output a 6-DOF compensation matrix including rigid compensation and deformation compensation.

[0037] Furthermore, the function expression of the 6-DOF compensation matrix is ​​configured as:

[0038]

[0039] Among them, M c represents the 6-DOF compensation matrix, exp represents the matrix exponential function, D r represents the rigid displacement component, Γ represents the Gamma function in matrix form, I represents the unit matrix, k represents the material deformation coefficient matrix, D drepresents the deformation displacement component, the rigid displacement component and the deformation displacement component are 6-dimensional vectors, β′ represents the preset acceleration suppression factor, |·| represents the Euclidean norm, E a represents the acceleration coupling error vector, erf represents the error function, γ′ represents the velocity sensitivity coefficient, E v represents the velocity coupling error vector.

[0040] Furthermore, the FPGA programmable register group of the sensor compensation circuit unit includes:

[0041] A temperature-voltage mapping lookup table is used to store the optimal bias voltage parameters at different temperatures;

[0042] a pulse width modulation generator that dynamically adjusts the signal sampling rate according to the compensation matrix;

[0043] The FPGA programmable register set is configured as:

[0044] When it is detected that the internal temperature suddenly changes and exceeds a threshold, the pre-compensated voltage value is loaded from the temperature-voltage mapping lookup table; the pulse width modulation generator is synchronously triggered to increase the sampling frequency to 3 times the normal value; and the sampling rate is restored to the base rate after the temperature stabilizes.

[0045] Furthermore, the temperature-voltage mapping lookup table adopts a quantized anti-interference storage structure, including:

[0046] Segmented quantum encoder is used to convert the temperature range [T min , T max ] is divided into N quantized intervals, and each interval is assigned a voltage value V n , and its calculation formula is configured as:

[0047] V n =V0·Γ(1+κ n ΔT)·erf(β n |ΔT|), where V0 is the reference voltage, κ n is the interval nonlinear coefficient, β n is the anti-interference gain, ΔT=T-T0 is the temperature change;

[0048] The aging compensation register is used to store the correction factor η(t) related to the sensor working time t and update the voltage value of each interval, wherein the calculation formula of the correction factor is configured as:

[0049] Where λ1 represents the preset attenuation factor, ζ and γ1 are the first and second coefficients related to the working time, i represents the imaginary unit, and the final compensated voltage value is:

[0050] Adjacent interval interpolator, used when the internal temperature T∈[T n , T n+1 ], use the interpolator to smooth the difference and calculate the output voltage V out , the calculation formula of the output voltage is configured as:

[0051] Where k is the interpolation steepness factor, which is used to control the smoothness of the interpolation.

[0052] Furthermore, the pulse width modulation generator executes a chaotic dynamic frequency modulation strategy, which includes:

[0053] Frequency chaos function: Set the basic sampling frequency f0 and calculate the frequency according to the acceleration error |E a | and temperature mutation rate The sampling frequency is dynamically adjusted, and the calculation formula of the sampling frequency is configured as:

[0054]

[0055] Where α2 is the acceleration coupling coefficient, E a is the acceleration error modulus, J0 is the zero-order Bessel function, β2 is the coupling coefficient of the temperature change rate, and γ2 is the temperature sensitivity coefficient;

[0056] Phase synchronization mechanism: When the temperature suddenly changes Threshold R th When , the chaotic disturbance term δf is injected to disturb the sampling frequency so that it can quickly adapt to temperature changes. The calculation formula of the chaotic disturbance term is configured as:

[0057] Where ζ1 and ω are the first and second disturbance factors related to time;

[0058] Oversampling protection: If the sampling frequency f>3f0, the anti-aliasing filter H(s) is activated to avoid signal aliasing, wherein the calculation formula of the anti-aliasing filter is configured as:

[0059] Cutoff frequency ω c =2πf0·erfc(k|ΔT|), where s represents the complex frequency variable in Laplace transform.

[0060] An intelligent sensor is applied to the above-mentioned intelligent sensor self-diagnosis optimization system, and the laser interference unit and the temperature detection unit are integrated in the intelligent sensor.

[0061] Beneficial effects of the present invention:

[0062] (1) With the help of the laser interference principle, the present invention can achieve nanometer-level displacement detection accuracy, significantly improving the measurement resolution compared to traditional capacitive or Hall-effect sensors, and is suitable for high-precision industrial measurement and equipment fine-tuning scenarios.

[0063] (2) The internal ambient temperature of the sensor is obtained through the temperature detection unit, and the displacement error caused by thermal expansion and contraction is corrected in combination with the thermal compensation model, thereby improving the reliability and stability of the system in a high-temperature fluctuating environment.

[0064] (3) This invention integrates edge computing chips to enable local sensor data processing and intelligent analysis, with high real-time performance, low latency, and self-learning capabilities. Compared with traditional processing architectures that rely on back-end servers, it has better self-adaptation and autonomous diagnostic capabilities.

[0065] (4) The introduction of a neural network-based target detection module can automatically identify the features of the reference workpiece within the sensor's field of view, correct spatial errors in real time, and effectively solve measurement deviation problems caused by structural vibration, installation errors, or workpiece offset.

[0066] (5) By dynamically loading the error compensation matrix through the FPGA register group, the original sensor signal can be adjusted in real time during the signal output stage, building a software and hardware collaborative compensation mechanism, significantly improving the overall measurement accuracy and stability.

[0067] (6) Compared with existing systems that are mostly static calibration or one-way compensation, the present invention establishes a closed-loop logic from physical perception to AI judgment and then to dynamic hardware compensation, realizing a true "self-diagnosis + self-adaptation" sensor optimization system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a structural diagram of the intelligent sensor self-diagnosis optimization system of the present invention;

[0069] Figure 2 It is a flow chart of the steps of the self-diagnosis optimization method of the intelligent sensor in the present invention.

[0070] Figure numerals: 1. Laser interference unit; 2. Temperature detection unit; 3. Edge computing unit; 31. Adjustment subunit; 32. Operation subunit; 33. Correction subunit; 4. Dynamic reference calibration unit; 41. Decomposition subunit; 42. Error calculation subunit; 43. Mapping subunit; 5. Sensor compensation circuit unit. DETAILED DESCRIPTION

[0071] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom," "top," "inner," and "outer" refer to directions toward or away from the geometric center of a particular component, respectively.

[0072] Example 1, with reference to Figure 1 , which is the first embodiment of the present invention, provides an intelligent sensor self-diagnosis optimization system, which can achieve high-precision dynamic error compensation and autonomous diagnosis optimization of intelligent sensors, including:

[0073] The laser interferometer unit 1 is integrated into the smart sensor and includes a laser emitter that emits visible red light and a photon detector array, which is used to obtain the displacement phase difference of the smart sensor;

[0074] The temperature detection unit 2 is integrated into the smart sensor and is used to detect the internal temperature of the smart sensor;

[0075] The edge computing unit 3 is communicatively connected to the laser interferometer unit 1 and the temperature detection unit 2, and has an integrated artificial intelligence processor chip for calculating the physical displacement of the smart sensor based on the displacement phase difference and the internal temperature;

[0076] A dynamic reference calibration unit 4 is communicatively connected to the edge computing unit 3 and is configured to utilize an object detection neural network to identify preset reference workpiece features within the field of view of the intelligent sensor on the production line, and to generate a compensation matrix for spatial error compensation based on the preset reference workpiece features and the physical displacement;

[0077] The sensor compensation circuit unit 5 includes an FPGA programmable register group, is communicatively connected to the dynamic reference calibration unit 4, and is used to dynamically configure the value of the FPGA programmable register group according to the compensation matrix to compensate the output signal of the intelligent sensor.

[0078] Working principle of embodiment 1:

[0079] The laser interferometer unit 1, integrated within the intelligent sensor, consists of a highly stable red laser (wavelength λ ≈ 650nm) and a two-dimensional photon detector array. The laser emitter irradiates the reflective surface of the mechanical structure under test. After interference, the detector array receives the coherent signal and extracts the tiny displacement of the target object through the phase difference of the interference fringes.

[0080] In order to compensate for structural expansion or refractive index changes caused by changes in the internal environment of the sensor, the system is embedded with a set of high-precision temperature sensors (accuracy ±0.01°C) to monitor the temperature status of the laser interference path and the structural shell in real time.

[0081] The Edge Computing Unit 3 integrates an AI inference chip (such as Cambricon or NVIDIA Jetson) that uses a neural network model to jointly infer the displacement phase difference and temperature data, outputting the corrected physical displacement. The Edge Computing Unit 3 also integrates a neural network model with strong real-time execution capabilities, supporting inference cycles of less than 50ms.

[0082] The dynamic reference calibration unit 4 deploys the YOLOv5 target detection model to analyze the production line images captured by the industrial camera in real time, automatically identify the preset reference workpiece features (such as marking points, edges, and graphics) in the field of view, and construct a three-dimensional spatial error vector based on the calculated physical displacement. Based on a series of error vectors obtained by continuous sampling, the unit uses the least squares method or Kalman filtering to construct a spatial error compensation matrix to represent the dynamic error weight relationship in different directions.

[0083] The sensor compensation circuit unit 5 includes an FPGA chip with a set of programmable registers. This register dynamically adjusts the gain parameters or displacement correction coefficients in the sensor output path based on the error compensation matrix. This process is written in real time by the edge computing unit 3 via the SPI communication interface, ultimately achieving hardware-level error compensation for the signal output.

[0084] Based on the above process, this embodiment completes a complete "perception-diagnosis-compensation" closed-loop process, achieving high-precision dynamic error compensation and autonomous diagnosis optimization of the smart sensor.

[0085] Example 2 is the second embodiment of the present invention. Unlike the previous embodiment, this embodiment mainly realizes effective suppression and precise compensation of interference signal drift in a high-temperature environment through the synergistic effect of an optical detection structure, a circuit-level dynamic response mechanism, and a disentanglement compensation algorithm based on physical modeling. Among them:

[0086] The photon detector array features a radially symmetric, hexagonal, close-packed structure. Its key advantages include: enhanced optical symmetry, which effectively offsets interference fringe shifts or spot drift caused by thermal flow; multipath redundant sampling, which allows for dynamic fitting of the interference field using edge multi-point detection data, improving spatial noise immunity; and enhanced phase recovery stability, which facilitates regional consistency determination in subsequent unwrapping algorithms. This array structure essentially "structurally averages" the interference caused by thermal distortion on a single path, creating a passive stabilization mechanism.

[0087] To suppress electrical noise and weak signal distortion caused by temperature rise, a variable-gain differential amplifier is installed between the center and edge detectors of the photon detector array. Its features include: dynamic gain adjustment: controlled by the adjustment subunit 31 of the edge computing unit 3, the gain of each amplifier path is adjusted in real time according to the temperature T; precise thermal drift resistance: the circuit level balances thermally induced gain inconsistencies in real time, eliminating the accumulation of small differences; and adaptive matching of interference signal strength: ensuring that the dynamic range of each path signal is symmetrical before entering the algorithm. This technical solution pre-compensates for dynamic signal imbalances caused by heat at the "circuit physical layer," contributing to thermal robustness at the front end of the signal chain.

[0088] The edge computing unit 3 includes:

[0089] an adjusting subunit 31 for dynamically adjusting a gain coefficient of a variable gain differential amplifier based on internal temperature data;

[0090] The operator unit 32 is used to perform a temperature gradient-based phase unwrapping operation on the displacement phase difference. Before phase unwrapping, the temperature-sensing model is used to adjust the phase mutation threshold of adjacent pixels to avoid misjudgment of phase jumps at high temperatures. After unwrapping, the phase difference is input into the following physical model to correct the nonlinear effect of temperature on the phase quantity:

[0091]

[0092] Wherein, represents the physical displacement, λ represents the laser wavelength of the laser transmitter, Δφ represents the displacement phase difference, α represents the phase compensation coefficient, erf represents the error function, β represents the temperature corresponding slope, T represents the internal temperature, T0 represents the preset reference temperature, γ represents the thermal expansion suppression factor, a represents the preset linear temperature drift coefficient, and b represents the preset nonlinear enhancement factor.

[0093] The correction subunit 33 is used to correct the displacement phase difference to eliminate thermo-optical distortion.

[0094] Working principle of embodiment 2:

[0095] The working process of the intelligent sensor self-diagnosis optimization system described in this embodiment is as follows:

[0096] Upon system startup, the laser interferometer unit 1 emits a stable red light beam, which generates an interference pattern through the workpiece or reference reflective surface. The photon detector array utilizes a radially symmetric hexagonal structure, consisting of a central detector and six edge detectors. This symmetrical arrangement minimizes interference pattern drift caused by uneven temperature expansion or optical axis offset. Multi-channel redundant signal sampling provides the data foundation for phase recovery and deviation estimation. A differential amplifier establishes signal comparison between the edge and central signals, creating a mechanism to mitigate common-mode thermal noise.

[0097] When the internal temperature of the sensor rises (for example, approaching 50°C), the temperature detection unit 2 monitors the current ambient temperature in real time. The adjustment subunit 31 in the edge computing unit 3 controls the gain coefficient of each variable-gain differential amplifier based on the T-T0 offset. This achieves the following functions: maintaining gain matching across channels to suppress electronic signal drift caused by temperature rise; optimizing the signal dynamic range to prevent excessive distortion of the interference fringe edge intensity due to thermal scattering; and providing raw data with a better signal-to-noise ratio for subsequent phase extraction.

[0098] After obtaining the amplified interference signal, the system performs the following process in the edge computing unit 3:

[0099] Pre-processing before unwrapping: The operator unit 32 first adjusts the phase jump threshold according to the temperature data to prevent non-real mutations induced by temperature (such as interference fringe blurring caused by heat) from being misinterpreted as phase jumps.

[0100] Phase unwrapping processing: Utilize the phase unwrapping algorithm based on minimum path error to expand the processing on the two-dimensional array and output a continuous phase distribution Δφ.

[0101] Thermo-optical distortion correction: The correction subunit inputs the unwrapped Δφ into the following temperature compensation model:

[0102]

[0103] In the laser interference path, temperature increases can cause nonlinear minor changes in the optical path difference due to changes in the air refractive index or component stress deformation, which manifests as phase drift. The phase compensation term α·erf(β(T-T0)) is used to compensate for the interference phase shift caused by temperature.

[0104] erf(x) is the error function, which is defined on (-∞, +∞), and the output value increases monotonically in the range of (-1, 1).

[0105] Under high temperature conditions, mechanical structures such as the laser housing and optical lens assembly will produce non-negligible thermal expansion effects, resulting in a decrease in interferometric sensitivity. The thermal expansion correction term exp(-γ(T-T0) 2 ) is used to describe the trend of nonlinear decrease in system sensitivity caused by temperature increase. 0＝0 The time value is 1, maintaining the ideal formula; as the square of the temperature difference increases, the exponential term decays rapidly, simulating a nonlinear decrease in sensitivity with temperature.

[0106] In the laser path or detection circuit, structural asymmetry or thermal expansion and contraction hysteresis effects can cause the overall measurement reference (zero point) to deviate. This function compensates for output zero-point offsets caused by asymmetric thermal stress or component aging. When |ΔT| = T - T0, the output changes approximately linearly near |ΔT| → 0, ensuring sensitive response to small drifts. When |ΔT| > 5°C, the compensation increases (nonlinear growth), ensuring steady-state protection under extreme thermal conditions. The zero-point drift term uses a rational function structure, ensuring smoothness and dynamic stability while avoiding high-order polynomial oscillation.

[0107] In summary, this model can accurately characterize the influence of multi-factor coupling under high temperature and output real and robust physical displacement. By using the above physical model, in the test at -20℃ to 80℃, the displacement error is controlled within ±1.2μm (traditional method ±20μm), and the phase unwrapping success rate is >99.5%.

[0108] After obtaining the corrected displacement result, the system compares the displacement result with the workpiece reference position dynamically identified on the production line, generates an error vector, and constructs an error compensation matrix. This compensation matrix is ​​input into the FPGA programmable register bank, enabling real-time hardware-level output signal adjustment and a closed-loop "perception → judgment → compensation" process throughout the entire process.

[0109] Beneficial effects of embodiment 2:

[0110] Unlike traditional temperature compensation, which only corrects the final output, this embodiment intervenes at the source of signal acquisition and actively controls the thermal distortion of the interference signal through a three-layer collaborative mechanism, providing stronger feedforward anti-interference capabilities.

[0111] This embodiment demonstrates outstanding cross-domain integration capabilities, encompassing multidisciplinary technology integration including optical detector layout optimization, circuit adjustable gain control, and AI-assisted disentanglement and thermal compensation modeling.

[0112] This embodiment has industrial practicality and scalability, can be deployed on standard FPGA and embedded edge computing platforms, and is easy to embed into existing industrial measurement and control systems.

[0113] Example 3, with reference to the figure, is the third embodiment of the present invention. Unlike the previous embodiment, this embodiment fully integrates the dual-stream neural network structure with the visual / depth fusion recognition mechanism, as well as the rigidity + deformation + dynamics multi-dimensional compensation logic, and can achieve the following:

[0114] In the dynamic reference calibration unit 4, the target detection neural network is a dual-stream fusion architecture, which includes:

[0115] The first branch processes visible light image features of production line workpieces. Its input is high-resolution visible light images captured by industrial cameras on the production line. Its network structure consists of a convolutional backbone (ResNet) and an attention module. It extracts 2D structural features such as workpiece surface texture, edges, and color, which are used for target framing and coarse positioning.

[0116] The second branch processes the depth phase point cloud generated by the laser interferometer unit 1. The input of the second branch is the depth phase point cloud (3D coordinates + phase attributes) calculated by the laser interferometer unit 1. The network structure is PointNet++ or a Transformer-based 3D feature extractor. Its function is to analyze the microstructure deformation and spatial posture of the workpiece, and its role is to provide depth reinforcement and precise positioning information.

[0117] The output of the dual-stream branch passes through the attention-guided feature fusion module to generate the spatial coordinates of the preset reference workpiece features. During the fusion process at the feature fusion gate, the spatial-channel attention weights are combined to achieve multi-scale saliency enhancement.

[0118] Preferably, the dynamic reference calibration unit 4 includes:

[0119] The decomposition subunit 41 is used to decompose the physical displacement into a rigid displacement component and a deformation displacement component, wherein the rigid displacement component D r =[t x , t y , t z ,ω x ,ω y ,ω z ] T , deformation displacement component D d , used to reflect the non-rigid offset caused by stress and material deformation;

[0120] The error calculation subunit 42 calculates the speed-acceleration coupling error of the production line conveyor belt based on the preset reference workpiece feature, where the speed coupling error Acceleration coupling error Error sources include dynamic factors such as belt acceleration and deceleration, inertial impact, and vibration interference;

[0121] The mapping subunit 43 connects the error calculation subunit 42 and the decomposition subunit 41 and is used to map the coupling error to the deformation displacement component through the Lie group SE (3) transformation model, and then output a 6-DOF compensation matrix including rigid compensation and deformation compensation.

[0122] Among them, the function expression of the 6-DOF compensation matrix is ​​configured as:

[0123]

[0124] Among them, Mc represents the 6-DOF compensation matrix, M c is written into the sensor compensation circuit (FPGA register group) to achieve 6-DOF hardware and software collaborative compensation. exp represents the matrix exponential function, D r represents the rigid displacement component, Γ represents the Gamma function in matrix form, I represents the unit matrix, k represents the material deformation coefficient matrix, D d represents the deformation displacement component, the rigid displacement component and the deformation displacement component are 6-dimensional vectors, β represents the preset acceleration suppression factor, |·| represents the Euclidean norm, E a represents the acceleration coupling error vector, erf represents the error function, γ represents the velocity sensitivity coefficient, E v represents the velocity coupling error vector.

[0125] in, The mathematical meaning of is Lie group mapping, which is used for rigid transformation. The 6-dimensional rigid displacement D is transformed by SE(3) Lie group index mapping. r Converts a 3×3 homogeneous transformation matrix that accurately describes the three-dimensional translation and rotation of the sensor installation location.

[0126] Γ(I+κD d ) represents the nonlinear deformation matrix for modeling the thermal stress of the material, κD d Realize anisotropic deformation coupling; exp(-β′||E a || 2 )·erf(γ′E v ) has the mathematical meaning of acceleration suppression and velocity drift correction to achieve dynamic compensation, where the acceleration term: exponential decay suppresses impact error, and the velocity term: error function compensates for velocity-related drift.

[0127] Specific application scenarios of the 6-DOF compensation matrix:

[0128] In the high-speed automobile body-in-white welding production line, the industrial robot's end effector is equipped with a laser probe for spot welding positioning. Due to frequent speed fluctuations, vibrations and thermal expansion and contraction of the workpiece in the acceleration section of the production line, the sensor output signal has significant dynamic drift. Traditional static compensation methods cannot meet the 0.01mm level accuracy requirements.

[0129] To this end, this system is based on the 6-DOF compensation matrix proposed in this invention, deployed in the welding main line sensor node, uses a two-stream neural network to identify the reference workpiece features, and combines the following parameters to complete a round of real-time compensation.

[0130] Input parameter: D r =[0.2, -0.1, 0.05, 0.01, -0.02, 0.005] T β′=0.12,Dd =[0.02, 0.01, -0.03, 0.001, 0.002, -0.001] T , E a =[1.2, -0.9, 0.6] T ,γ′=0.85,E v =[0.12, -0.08, 0.05] T ;

[0131]

[0132] Calculation process:

[0133] Rigidity Matrix The calculation result is:

[0134]

[0135] Nonlinear deformation matrix Γ(I+κD d ) is calculated as:

[0136]

[0137] The calculation result of the acceleration attenuation term is:

[0138] exp(-β'||E a || 2 )=exp(-0.12×2.61)≈0.382;

[0139] The velocity error term is calculated as:

[0140] erf(γ′E v )=erf(0.85·[0.12,-0.08,0.05])≈[0.102,-0.068,0.043];

[0141] Then the product of the acceleration attenuation term and the velocity error term is:

[0142] exp(-β'||E a || 2 )·erf(γ′E v )≈[0.039,-0.026,0.016];

[0143] The final 6-DOF compensation matrix M c The calculation result is:

[0144]

[0145] According to the above calculation results, the compensation effect table of the compensation matrix is ​​obtained as follows:

[0146]

[0147] Table 1

[0148] As can be seen from Table 1, through calculation and compensation, the system can significantly improve position and angle accuracy, reduce FPGA resource consumption by about 21%, and maintain stable real-time compensation performance under high dynamic conditions.

[0149] Working principle of embodiment 3:

[0150] The industrial camera on the production line continuously collects images of the workpiece surface and inputs them into the first branch of the neural network. The laser interferometer unit 1 measures the displacement phase difference in real time, and the edge computing unit 3 converts it into a 3D point cloud and inputs it into the second branch of the network.

[0151] The first branch extracts image texture features (position, boundary, shape); the feature fusion module aligns and weights the two based on the attention mechanism; output: obtains the three-dimensional spatial coordinates of the preset reference workpiece in the sensor's field of view, and compares them with historical calibration values ​​to generate an error signal.

[0152] The current physical displacement of the sensor is obtained through laser interferometry and temperature compensation formula (see Example 2). The system decomposes the displacement into a rigid displacement component and a deformation displacement component: the deformation component reflects the non-rigid deformation of the workpiece due to heat or mechanical stress.

[0153] Dynamic disturbance extraction: Real-time perception of speed and acceleration changes on the production line; inference of dynamic offset trends caused by inertia, vibration, and flexible tooling.

[0154] The system generates the compensation matrix through the following model:

[0155]

[0156] The matrix is ​​transmitted to the sensor compensation circuit unit; the FPGA configuration register adjusts the output signal in real time to achieve dynamic correction of posture accuracy.

[0157] Example 4, with reference to the figures, is the fourth embodiment of the present invention. Unlike the previous embodiment, this one provides an FPGA-programmable register bank that integrates a temperature-voltage mapping lookup table, a pulse-width modulation generator, and a chaotic dynamic frequency modulation strategy. This optimizes the smart sensor's self-diagnosis and compensation capabilities for dynamic temperature fluctuations and aging. This embodiment dynamically adjusts the sensor's signal sampling rate and voltage compensation based on ambient temperature changes and operating conditions, ensuring measurement accuracy and stability.

[0158] Example 4 Working Principle:

[0159] The FPGA programmable register bank contains a temperature-voltage mapping lookup table that stores the optimal bias voltage parameters at different temperatures. This lookup table uses a quantized anti-interference storage structure and mainly includes the following parts:

[0160] Segmented quantum encoder is used to convert the temperature range [T min , T max ] is divided into N quantized intervals, and each interval is assigned a voltage value V n , and its calculation formula is configured as:

[0161] V n =V0·Γ(1+κ n ΔT)·erf(β n |ΔT|), where V0 is the reference voltage, κ n is the interval nonlinear coefficient, β n is the anti-interference gain, ΔT=T-T0 is the temperature change;

[0162] The aging compensation register is used to store the correction factor η(t) related to the sensor working time t and update the voltage value of each interval. The calculation formula of the correction factor is configured as:

[0163] Where λ1 represents the preset attenuation factor, ζ and γ1 are the first and second coefficients related to the working time, i represents the imaginary unit, and the final compensated voltage value is:

[0164] Adjacent interval interpolator, used when the internal temperature T∈[T n , T n+1 ], use the interpolator to smooth the difference and calculate the output voltage V out , the calculation formula of the output voltage is configured as:

[0165] Where k is the interpolation steepness factor, which is used to control the smoothness of the interpolation.

[0166] When the internal temperature is detected to exceed the set threshold, the FPGA will perform the following operations:

[0167] Load the pre-compensated voltage value from the temperature-voltage mapping lookup table and apply it to the sensor;

[0168] Start the pulse width modulation (PWM) generator to increase the signal sampling frequency to three times the normal value to enhance the dynamic response capability under temperature fluctuations;

[0169] After the temperature stabilizes, the system will restore the sampling frequency to the baseline value.

[0170] Preferably, the pulse width modulation generator implements a chaotic dynamic frequency modulation strategy, which includes:

[0171] Frequency chaos function: Set the basic sampling frequency f0 and calculate the frequency according to the acceleration error |E a | and temperature mutation rate Dynamically adjust the sampling frequency. The calculation formula for the sampling frequency is configured as:

[0172]

[0173] Where α2 is the acceleration coupling coefficient, E a is the acceleration error modulus, J0 is the zero-order Bessel function, β2 is the coupling coefficient of the temperature change rate, and γ2 is the temperature sensitivity coefficient;

[0174] Phase synchronization mechanism: When the temperature suddenly changes Threshold R th When , the chaotic disturbance term δf is injected to disturb the sampling frequency so that it can quickly adapt to temperature changes. The calculation formula of the chaotic disturbance term is configured as:

[0175] Where ζ1 and ω are the first and second disturbance factors related to time;

[0176] Oversampling protection: If the sampling frequency f>3f0, the anti-aliasing filter H(s) is activated to avoid signal aliasing. The calculation formula of the anti-aliasing filter is configured as follows:

[0177] Cutoff frequency ω c =2πf0·erfc(κ|ΔT|), where s represents the complex frequency variable in Laplace transform.

[0178] Workflow summary:

[0179] 1. Temperature change monitoring: FPGA continuously monitors internal temperature fluctuations;

[0180] 2. Temperature mutation detection: When the temperature mutation exceeds the threshold, the corresponding compensation voltage value is loaded and the PWM generator is triggered;

[0181] 3. Dynamic sampling adjustment: Dynamically adjust the sampling frequency according to temperature and acceleration errors to ensure that the impact of temperature fluctuations on measurement is minimal;

[0182] 4. Signal compensation: Real-time update of temperature-related voltage values ​​and execution of chaotic dynamic frequency modulation to maintain signal accuracy;

[0183] 5. Restore after temperature stabilizes: When the temperature returns to a stable state, restore the reference sampling frequency and voltage compensation.

[0184] Beneficial effects of Example 4:

[0185] Improve the ability to respond to sudden temperature changes: By dynamically adjusting the sampling frequency and voltage compensation, the system can effectively cope with high-frequency temperature fluctuations and reduce the impact of temperature changes on measurement accuracy.

[0186] Strong anti-interference capability: The quantized anti-interference storage structure and chaotic dynamic frequency modulation strategy effectively improve the robustness of the system in noisy and interference environments.

[0187] Aging compensation: Real-time update of voltage compensation through aging compensation registers ensures the stability and accuracy of the sensor under long-term operation.

[0188] Dynamic adaptability: The system can dynamically adjust the sampling frequency according to acceleration and temperature changes to adapt to complex industrial environments.

[0189] The design of this embodiment effectively enhances the self-diagnosis and optimization capabilities of the smart sensor, and is particularly suitable for high-temperature environments and those with large acceleration fluctuations. By refining the working principle and application details, we can better understand the system's dynamic compensation and enhanced adaptability.

[0190] An intelligent sensor is applied to the above-mentioned intelligent sensor self-diagnosis optimization system, and the laser interference unit 1 and the temperature detection unit 2 are integrated in the intelligent sensor.

[0191] An intelligent sensor self-diagnosis collaborative method is applied to the above intelligent sensor self-diagnosis optimization system, referring to Figure 2 ,include:

[0192] Step S1, the laser interferometer unit 1 obtains the displacement phase difference of the smart sensor;

[0193] Step S2, the temperature detection unit 2 detects the internal temperature of the smart sensor;

[0194] Step S3: The edge computing unit 3 calculates the physical displacement of the smart sensor based on the displacement phase difference and the internal temperature;

[0195] Step S4: The dynamic reference calibration unit 4 uses the target detection neural network to identify the preset reference workpiece features within the field of view of the intelligent sensor on the production line, and generates a compensation matrix for spatial error compensation based on the preset reference workpiece features and the physical displacement;

[0196] In step S5 , the sensor compensation circuit unit 5 dynamically configures the values ​​of the FPGA programmable register group according to the compensation matrix to compensate the output signal of the smart sensor.

[0197] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention are within the scope of protection of the present invention.

Claims

1. An intelligent sensor self-diagnosis optimization system, characterized in that: include: A laser interference unit (1) is integrated into the smart sensor and comprises a laser emitter emitting visible red light and a photon detector array, and is used to obtain the displacement phase difference of the smart sensor; A temperature detection unit (2), integrated into the smart sensor, for detecting the internal temperature of the smart sensor; an edge computing unit (3), communicatively connected to the laser interference unit (1) and the temperature detection unit (2), and having an integrated artificial intelligence processor chip, for calculating the physical displacement of the smart sensor based on the displacement phase difference and the internal temperature; a dynamic reference calibration unit (4), communicatively connected to the edge computing unit (3), for identifying preset reference workpiece features within the field of view of the smart sensor on the production line using a target detection neural network, and generating a compensation matrix for spatial error compensation based on the preset reference workpiece features and the physical displacement; A sensor compensation circuit unit (5) includes an FPGA programmable register group, is communicatively connected to the dynamic reference calibration unit (4), and is used to dynamically configure the value of the FPGA programmable register group according to the compensation matrix to compensate the output signal of the intelligent sensor.

2. The intelligent sensor self-diagnosis and optimization system according to claim 1, characterized in that: The photon detector array is a radially symmetrical hexagonal close-packed structure, with a variable gain differential amplifier provided between the central detector and the edge detector; The edge computing unit (3) comprises: an adjusting subunit (31), configured to dynamically adjust a gain coefficient of the variable gain differential amplifier based on the internal temperature data; An operation subunit (32), configured to perform a temperature gradient-based phase unwrapping operation on the displacement phase difference; The correction subunit (33) is used to correct the displacement phase difference to eliminate thermo-optical distortion.

3. The intelligent sensor self-diagnosis and optimization system according to claim 2, characterized in that: The calculation formula of the physical displacement is configured as: Wherein, represents the physical displacement, λ represents the laser wavelength of the laser emitter, Δφ represents the displacement phase difference, α represents the phase compensation coefficient, erf represents the error function, β represents the temperature corresponding slope, T represents the internal temperature, T0 represents the preset reference temperature, γ represents the thermal expansion suppression factor, a represents the preset linear temperature drift coefficient, and b represents the preset nonlinear enhancement factor.

4. The intelligent sensor self-diagnosis and optimization system according to claim 1, characterized in that: In the dynamic reference calibration unit (4), the target detection neural network is a dual-stream fusion architecture, and the dual-stream fusion architecture includes: The first branch processes the visible light image features of the production line workpieces; The second branch processes the depth phase point cloud generated by the laser interferometer unit (1); The output of the dual-stream branch is passed through an attention-guided feature fusion module to generate the spatial coordinates of the preset reference workpiece features.

5. The intelligent sensor self-diagnosis optimization system according to claim 4, characterized in that: The dynamic reference calibration unit (4) comprises: A decomposition subunit (41) is used to decompose the physical displacement into a rigid displacement component and a deformation displacement component; an error calculation subunit (42) for calculating a speed-acceleration coupling error of a production line conveyor belt based on the preset reference workpiece feature; A mapping subunit (43) is connected to the error calculation subunit (42) and the decomposition subunit (41), and is used to map the coupling error to the deformation displacement component through the Lie group SE (3) transformation model, and then output a 6-degree-of-freedom compensation matrix including rigid compensation and deformation compensation.

6. The intelligent sensor self-diagnosis and optimization system according to claim 5, characterized in that: The functional expression of the 6-DOF compensation matrix is ​​configured as: Among them, M c represents the 6-DOF compensation matrix, exp represents the matrix exponential function, D r represents the rigid displacement component, Γ represents the Gamma function in matrix form, I represents the unit matrix, κ represents the material deformation coefficient matrix, D d represents the deformation displacement component, the rigid displacement component and the deformation displacement component are 6-dimensional vectors, β′ represents the preset acceleration suppression factor, |·| represents the Euclidean norm, E a represents the acceleration coupling error vector, erf represents the error function, γ′ represents the velocity sensitivity coefficient, E v represents the velocity coupling error vector.

7. The intelligent sensor self-diagnosis and optimization system according to claim 3, characterized in that: The FPGA programmable register group of the sensor compensation circuit unit includes: A temperature-voltage mapping lookup table is used to store the optimal bias voltage parameters at different temperatures; a pulse width modulation generator that dynamically adjusts the signal sampling rate according to the compensation matrix; The FPGA programmable register set is configured as: When it is detected that the internal temperature suddenly changes and exceeds a threshold, the pre-compensated voltage value is loaded from the temperature-voltage mapping lookup table; the pulse width modulation generator is synchronously triggered to increase the sampling frequency to 3 times the normal value; and the sampling rate is restored to the base rate after the temperature stabilizes.

8. The intelligent sensor self-diagnosis and optimization system according to claim 7, characterized in that: The temperature-voltage mapping lookup table adopts a quantized anti-interference storage structure, including: Segmented quantum encoder is used to convert the temperature range [T min , T max ] is divided into N quantized intervals, and each interval is assigned a voltage value V n , and its calculation formula is configured as: V n =V0·Γ(1+k n ΔT)·erf(β n |ΔT|), where V0 is the reference voltage, κ n is the interval nonlinear coefficient, β n is the anti-interference gain, ΔT=T-T0 is the temperature change; The aging compensation register is used to store the correction factor η(t) related to the sensor working time t and update the voltage value of each interval, wherein the calculation formula of the correction factor is configured as: Where λ1 represents the preset attenuation factor, ζ and γ1 are the first and second coefficients related to the working time, i represents the imaginary unit, and the final compensated voltage value is: Adjacent interval interpolator, used when the internal temperature T∈[T n , T n+1 ], use the interpolator to smooth the difference and calculate the output voltage V out , the calculation formula of the output voltage is configured as: Where k is the interpolation steepness factor, which is used to control the smoothness of the interpolation.

9. The intelligent sensor self-diagnosis and optimization system according to claim 7, characterized in that: The pulse width modulation generator implements a chaotic dynamic frequency modulation strategy, which includes: Frequency chaos function: Set the basic sampling frequency f0 and calculate the frequency according to the acceleration error |E a | and temperature mutation rate The sampling frequency is dynamically adjusted, and the calculation formula of the sampling frequency is configured as: Where α2 is the acceleration coupling coefficient, E a is the acceleration error modulus, J0 is the zero-order Bessel function, β2 is the coupling coefficient of the temperature change rate, and γ2 is the temperature sensitivity coefficient; Phase synchronization mechanism: When the temperature suddenly changes When , the chaotic disturbance term δf is injected to disturb the sampling frequency so that it can quickly adapt to temperature changes. The calculation formula of the chaotic disturbance term is configured as: Where ζ1 and ω are the first and second disturbance factors related to time; Oversampling protection: If the sampling frequency f>3f0, the anti-aliasing filter H(s) is activated to avoid signal aliasing, wherein the calculation formula of the anti-aliasing filter is configured as: Cutoff frequency ω c =2πf0·erfc(κ|ΔT|), where s represents the complex frequency variable in Laplace transform.

10. An intelligent sensor, applied to the intelligent sensor self-diagnosis optimization system according to any one of claims 1 to 9, characterized in that: The laser interference unit (1) and the temperature detection unit (2) are integrated therein.

Images