Dynamic profile identification method based on real-time measurement of flexible array pressure sensor

Abstract

The present application belongs to the technical field of automobile collision analysis, and particularly relates to a dynamic characteristic profile identification method based on real-time measurement of a flexible array pressure sensor. The sensor is first arranged on the front of a dummy and bound with a safety belt. During a collision test, pressure data is sampled through a parallel hierarchical staggered sampling architecture, and a collection result is obtained through time cross-sampling combination. Then, the result is sequentially subjected to signal amplification, noise removal, data correction and missing supplement. Subsequently, a dummy force surface profile visualization interface is constructed, the pressure mean value in a preset time window is calculated as a display threshold, and the interface parameters are updated. Then, the sensor discrete pressure points in the interface are interpolated and smoothed, and a binary image is obtained through original image processing. Finally, a profile extraction algorithm is called to complete core profile extraction and post-processing, and a safety belt acting dummy force profile visualization image is obtained. The present application can solve the problems of insufficient monitoring precision of safety belt on dummy force distribution and low data analysis visualization degree in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of automotive collision analysis technology, and particularly relates to a dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor. Background Technology

[0002] Automotive crash testing is a core method for evaluating vehicle passive safety performance and guiding vehicle body structure optimization and restraint system design. The test results directly affect the vehicle's ability to protect occupants in actual collisions. In crash tests, anthropomorphic dummies, as substitutes for human physiological characteristics and mechanical responses, simulate the occupant's posture changes and force states at the moment of impact. Seat belts, as the primary and most fundamental restraint device during a collision, have their contact surface with the dummy's body, including force distribution, load transfer efficiency, and dynamic response characteristics. These factors are crucial for assessing the effectiveness of seat belt restraint and determining the risk of injury to critical areas such as the occupant's chest and abdomen. Therefore, accurate capture and in-depth analysis of the dummy's force profile under seat belt action during crash tests are key technical supports for optimizing seat belt structural parameters (such as webbing width and locking timing) and improving occupant protection levels.

[0003] Currently, in the force monitoring of crash test dummies in automobiles, the industry generally adopts a technical solution of "single-point pressure sensors + data acquisition device" to acquire force data under the action of seat belts. Specifically, technicians pre-position discrete single-point pressure sensors on key parts of the dummy that directly contact the seat belt, such as the chest and abdomen. When a collision occurs, the sensors collect the instantaneous pressure values ​​at the contact points in real time. After preliminary filtering by the data acquisition device (usually a simple low-pass filter), the raw pressure data is transmitted to the back-end data processing system. Subsequently, by statistically analyzing the pressure values ​​at discrete points (such as calculating peak pressure and average pressure), and combining them with the dummy's posture parameters, a preliminary assessment of the seat belt's restraint effect is made. In terms of data visualization, existing technologies mostly present data in the form of numerical tables or single-dimensional pressure-time curves, relying on testers to interpret the correlation and force patterns of the data based on their experience, and then form an evaluation conclusion on the seat belt's performance.

[0004] Therefore, existing technologies suffer from insufficient accuracy in monitoring the force distribution of the seat belt on the dummy and low levels of data analysis visualization. Conventional single-point pressure sensors struggle to capture the continuous pressure distribution at the contact surface between the seat belt and the dummy. They also lack multi-level filtering mechanisms for power frequency interference and outliers, affecting data accuracy. Furthermore, collision energy analysis and force visualization do not form a complete closed loop, relying on manual data interpretation. Summary of the Invention

[0005] The technical problem solved by this invention is to provide a dynamic feature contour recognition method based on real-time measurement of a flexible array pressure sensor, so as to solve the problems of insufficient monitoring accuracy of the force distribution of the dummy by the seat belt and low degree of data analysis visualization in the prior art.

[0006] The basic solution provided by this invention is a dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor, comprising:

[0007] S1: Flexible array pressure sensors and safety belts are sequentially deployed on the front of the dummy. During the crash test, the flexible array pressure sensors sample pressure data through a parallel hierarchical staggered sampling architecture, and the sampling results are merged through time cross-sampling to obtain the pressure acquisition results.

[0008] S2: The pressure acquisition results are sequentially processed by signal amplification, signal noise identification and removal, data correction and missing data supplementation to obtain the processed pressure acquisition results;

[0009] S3: Collect raw image data of the dummy under the action of the seat belt during the crash test, construct a visualization interface of the dummy's force surface contour, and calculate the average pressure value in the pressure acquisition results within the preset time window in real time as the current display threshold, and update the parameters of the dummy's force surface contour visualization interface synchronously.

[0010] S4: Perform interpolation smoothing on the discrete pressure points of the pressure sensor in the visualization interface of the dummy's force surface contour; perform grayscale processing, image noise filtering, and image binarization on the original image data in the visualization interface of the dummy's force surface contour to obtain a binary image.

[0011] S5: Call the preset contour extraction algorithm to extract the core contour of the binary image and perform contour post-processing to obtain the final visualization image of the force contour of the seat belt action dummy.

[0012] Furthermore, S1 includes:

[0013] S1-1: Install a flexible array of pressure sensors on the front of the dummy, monitor the initial force information of the flexible array of pressure sensors, assign a unique identifier to each pressure sensor, and establish the correspondence between the pressure sensor and its physical position on the front of the dummy through a signal acquisition system.

[0014] S1-2: Ensure the dummy is fully in contact with and secured to the seatbelt, keeping the dummy and seatbelt stationary;

[0015] S1-3: Perform zero-point calibration, full-scale calibration, and linear calibration on the flexible array pressure sensor, and adjust the dummy and seat belt according to the calibration anomaly signal.

[0016] Furthermore, S1 also includes:

[0017] S1-4: Construct a virtual model of the visualization system for seat belt effect dummy crash testing;

[0018] S1-5: Conduct collision tests using a parallel hierarchical interleaved sampling architecture, including sampling the same signal through M ADCs with a fixed phase difference, with each ADC operating at a frequency of 1 / M of the total frequency;

[0019] S1-6: The sampling results are merged through time cross-sampling, including controlling the phase difference of each clock through a programmable phase shifter circuit, receiving ADC output data in DDR mode, and performing timing synchronization with FPGA;

[0020] S1-7: Synchronously monitor the seat belt slippage during the crash test, calculate the seat belt tension distribution unevenness, and monitor the seat belt locking trigger time;

[0021] S1-8: The frequency domain analysis method is adopted to separate the high-frequency seat belt load signal through low-pass filtering; the gas expansion dynamics model and finite element simulation are established by combining the airbag pressure sensor data; by using the time difference between the seat belt and the airbag triggering, the force of the steering wheel and airbag on the seat belt is screened out in the time domain to distinguish the force action stage, and the final pressure acquisition result is generated.

[0022] Furthermore, S1-8 specifically refers to:

[0023] Frequency domain analysis was used to separate the high-frequency seatbelt load signal through low-pass filtering. ;

[0024] A gas collision dynamics model and finite element simulation were established based on airbag pressure sensor data. The expression is as follows:

[0025]

[0026] in, This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. This represents the gas leakage correction factor; express The mass of gas inside the airbag at all times. Represents the gas constant; express The temperature of the gas inside the airbag at all times. express Real-time volume of the airbag;

[0027] The expression for the time difference between seatbelt and airbag deployment is:

[0028]

[0029] in, This indicates the time difference between the seatbelt and the airbag deployment. Indicates the airbag trigger time. Indicates the moment the seatbelt is triggered;

[0030] During the time-domain force differentiation phase, the forces exerted by the steering wheel and airbag on the seatbelt are filtered out, generating the final pressure acquisition result, expressed as:

[0031]

[0032] in, This indicates the final pressure acquisition result. This represents the restored time-domain high-frequency seatbelt load signal. This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. Indicates the start time of the crash test. Indicates the airbag trigger time. Indicates the end time of the crash test. This indicates the stage where the seatbelt is in full effect. This indicates the combined action phase of the seatbelt and airbag.

[0033] Furthermore, S2 includes:

[0034] S2-1: The pressure acquisition results are synchronously amplified by an instrumentation amplifier and an operational amplifier, and then converted into a digital pressure signal;

[0035] S2-2: A second-order active low-pass filter is used to suppress 50Hz power frequency interference, and the pressure digital signal is sequentially processed by moving average filtering, Kalman filtering, and median filtering to obtain the preprocessed pressure digital signal.

[0036] S2-3: Real-time monitoring of the zero point of the flexible array pressure sensor, calculation of correction values ​​for the pressure digital signal based on a preset temperature-zero point drift third-order polynomial model, and elimination of abnormal drift points through a sliding window calculation.

[0037] S2-4: Real-time monitoring of the damaged pressure sensor at the damaged location in the flexible array pressure sensor, calling the undamaged pressure sensors around the damaged location for average value processing, and automatically filling the pressure data after processing with the pressure data of the damaged pressure sensor.

[0038] Furthermore, S3 includes:

[0039] S3-1: Convert the pressure digital signal into pressure value, combine the physical position of the pressure sensor on the dummy with the mapping table of the unique identifier of the pressure sensor, assign the pressure value of each pressure sensor to the corresponding position of the virtual model of the visualization system, and use the 3D rendering engine to generate the force surface contour map of the dummy.

[0040] S3-2: Calculate the average pressure value in the pressure acquisition results within the preset time window in real time, use it as the current display threshold, and update the parameters of the dummy force surface contour visualization interface simultaneously.

[0041] Furthermore, S4 includes:

[0042] S4-1: In the visualization interface of the dummy's force surface contour, the discrete pressure points of the sampled pressure sensor are used to generate a smooth curve by optimizing the control points using a cubic B-spline interpolation algorithm to eliminate contour jaggedness.

[0043] S4-2: In the visualization interface of the dummy's stress surface contour, the original image data is converted into a grayscale image, Gaussian blur algorithm is applied to suppress noise, and adaptive thresholding method is used to convert the grayscale image into a binary image, fill the contour depression and eliminate burrs.

[0044] Furthermore, S5 includes:

[0045] S5-1: Call the OpenCV cv2.findContours() function, select the contour retrieval mode and contour approximation method, scan the preprocessed binarized image, output the contour point set and hierarchical information, and obtain the initial extracted dummy force surface contour.

[0046] S5-2: Call the OpenCV function cv2.contourArea() to calculate the contour area, and preset an area threshold to filter out small noise points. The expression is:

[0047]

[0048] The set of pixel coordinates of the contour is as follows: ;

[0049] A preset area threshold is established based on the minimum contact area of ​​the dummy's seat belt. ,when The outlines were identified as noise and removed.

[0050] S5-3: Real-time calculation of the contact area between the seat belt and the dummy, expressed as:

[0051]

[0052] in, for Rate of change of contact area at any time; for Actual contact area at any given moment; The initial contact area before the collision;

[0053] S5-4: Establish a pressure level-contact area change rate assessment table, based on real-time pressure values ​​and contact area change rates. Determine the risk level;

[0054] S5-5: Synchronously bind the dummy's stress surface contour cloud map with time series data, transmit it to the virtual model of the visualization system to establish a time axis, drag the time axis to display the stress contour back at any time, and mark and display the pressure data at the current time;

[0055] S5-6: Calculate the energy absorption distribution in the contact area of ​​the dummy, establish a histogram of regional energy proportion, calculate the energy concentration, and obtain the energy analysis results.

[0056] The principle and advantages of this invention are as follows: The technical solution of this application relies on a flexible array pressure sensor to realize the dynamic feature contour recognition of the dummy under force. During testing, a flexible array pressure sensor was worn on a dummy. Pressure signals were acquired using a parallel hierarchical interleaved sampling architecture. This architecture utilizes multi-channel ADC fixed-phase-difference sampling and FPGA timing synchronization to achieve equivalent high-speed data acquisition. The acquired analog signals were synchronously amplified by instrumentation amplifiers and operational amplifiers, converted into digital signals, and then subjected to a second-order active low-pass filter to suppress power frequency interference. Moving average filtering, Kalman filtering, and median filtering were used to remove signal noise. To address sensor zero-point drift, a temperature-zero-point drift model was established for correction and outlier removal. When a local sensor was damaged, surrounding data was used to supplement missing information. Subsequently, the mean or median of the most recent time window data was calculated to dynamically adjust the display threshold. Cubic B-spline interpolation was used to eliminate jagged edges for discrete pressure points. After image preprocessing, a dedicated function was called to extract the contour. Noise points were then filtered based on geometric features. A three-dimensional contour was generated by combining the sensor's three-dimensional coordinates and pressure data. Simultaneously, frequency domain analysis and timing differentiation were used to filter out the forces exerted by the steering wheel and airbag on the seat belt, ultimately achieving accurate dynamic feature contour recognition.

[0057] The advantages are: it effectively solves the problems of insufficient accuracy in monitoring the force distribution of dummy crashes by seat belts and low visualization of data analysis in existing technologies. Through the layout design of flexible array pressure sensors, combined with multi-level amplification and filtering technology, the accuracy and range coverage of pressure monitoring are greatly improved, ensuring accurate capture of the dummy's force situation in crash tests; noise processing and data correction mechanisms ensure data reliability and reduce the impact of interference and missing data on analysis results; dynamic threshold adjustment and contour processing technology avoid abnormal contour display, and the generated three-dimensional contour can intuitively present the force distribution, making it easy to clearly observe the force differences in different parts of the dummy; at the same time, it can also realize real-time monitoring of seat belt-related parameters and screening of interference forces. Combined with injury risk assessment tables and energy warning mechanisms, it forms a complete process system from data acquisition to analysis and evaluation, improving the intelligence and visualization level of crash test analysis and providing strong data support for optimizing the protective performance of automotive seat belts. Attached Figure Description

[0058] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0059] The following detailed description illustrates the specific implementation method:

[0060] The implementation examples are basically as follows Figure 1 As shown: A dynamic feature contour recognition method based on real-time measurement using a flexible array pressure sensor, including:

[0061] S1: A flexible array pressure sensor and a safety belt are sequentially deployed on the front of the dummy. During the crash test, the flexible array pressure sensor samples pressure data through a parallel hierarchical interleaved sampling architecture, and the sampling results are merged through time-cross sampling to obtain the pressure acquisition result; wherein, S1 includes:

[0062] S1-1: Install a flexible array of pressure sensors on the front of the dummy, monitor the initial force information of the flexible array of pressure sensors, assign a unique identifier to each pressure sensor, and establish the correspondence between the pressure sensor and its physical position on the front of the dummy through a signal acquisition system.

[0063] S1-2: Ensure the dummy is fully in contact with and secured to the seatbelt, keeping the dummy and seatbelt stationary;

[0064] S1-3: Perform zero-point calibration, full-scale calibration, and linear calibration on the flexible array pressure sensor, and adjust the dummy and seat belt according to the calibration anomaly signal.

[0065] In this embodiment, the flexible array pressure sensor refers to a pressure sensor made of flexible material and arranged in an array, specifically using... Layout, minimum resolution is Pressure measurement range is precision Each pressure sensor is equipped with an independent temperature compensation module, with a compensation range of [missing information]. Compensation accuracy .

[0066] The pressure sensors are attached to the chest, abdomen, and other key areas on the front of the dummy that are in direct contact with the seat belt. During installation, ensure that the sensors are completely attached to the dummy's skin and that the seat belt webbing is in full contact with the pressure sensor array, without any gaps or wrinkles.

[0067] Simultaneously, a unique identifier ID is assigned to each pressure sensor, such as an 8-bit binary address code, and a signal transmission link is constructed using a CAN bus or RS485 protocol; and the correspondence between physical location and electrical signal is established through a signal acquisition system. Specifically, the three-dimensional coordinates of each pressure sensor are entered into the system configuration file directory (with the midpoint of the dummy's sternum as the origin, an xyz coordinate system is established), forming an ID-coordinate mapping table to ensure that subsequent data can be traced back to the specific contact location.

[0068] For the calibration process of pressure sensors, the first step is to prepare the calibration environment and standard source. The calibration environment is... relative humidity Calibration is performed under standard conditions, using three levels of standard weights as the pressure source, such as 20kN, 44kN, and 80kN, and the accuracy class of the calibration device is not lower than 0.01.

[0069] Next, zero-point calibration and full-scale and linear calibration are performed. Zero-point calibration involves acquiring the output signal value of each pressure sensor under no-load conditions and recording it as the zero-point reference. If the zero-point reference value exceeds the preset threshold, it is corrected by the hardware zeroing circuit. Full-scale and linear calibration involves sequentially loading three levels of standard weights, acquiring the corresponding output signal values, and using the least squares method to fit the linear relationship between the pressure value and the output signal to establish a calibration equation for full-scale and linear calibration.

[0070] Finally, anomaly investigation was conducted. If signal jumps or linear deviations occurred during the calibration process, the abnormal pressure sensor was located using the ID-coordinate mapping table. The cause was investigated to determine if it was due to seat belt slippage, dummy tilt, or improper sensor fit. After adjustment, the calibration was repeated.

[0071] In addition, S1 also includes:

[0072] S1-4: Construct a virtual model of the visualization system for seat belt effect dummy crash testing;

[0073] S1-5: Conduct collision tests using a parallel hierarchical interleaved sampling architecture, including sampling the same signal through M ADCs with a fixed phase difference, with each ADC operating at a frequency of 1 / M of the total frequency;

[0074] S1-6: The sampling results are merged through time cross-sampling, including controlling the phase difference of each clock through a programmable phase shifting circuit (such as an LC network), receiving ADC output data in DDR mode, and performing timing synchronization with FPGA;

[0075] S1-7: Synchronously monitor the seat belt slippage during the crash test, calculate the seat belt tension distribution unevenness, and monitor the seat belt locking trigger time;

[0076] S1-8: The frequency domain analysis method is adopted to separate the high-frequency seat belt load signal through low-pass filtering; the gas expansion dynamics model and finite element simulation are established by combining the airbag pressure sensor data; by using the time difference between the seat belt and the airbag triggering, the force of the steering wheel and airbag on the seat belt is screened out in the time domain to distinguish the force action stage, and the final pressure acquisition result is generated.

[0077] In this embodiment, the construction process of the virtual model of the visualization system involves using HyperMesh software to create a finite element model of a dummy, incorporating muscle-skeleton coupling characteristics. The model includes biomechanical parameters of key parts such as the sternum, ribs, and abdomen (e.g., bone elastic modulus of 20 GPa, muscle Poisson's ratio of 0.45), with a mesh generation accuracy of 2 mm × 2 mm, matching the resolution of the flexible sensor. The established model presets key biomechanical index thresholds, such as chest compression. 60mm, peak abdominal pressure 65kN, the pressure corresponding to the risk threshold of rib fracture 70kN, etc., will serve as the basis for subsequent risk assessment.

[0078] Subsequently, a parallel hierarchical interleaved sampling architecture is used for sampling, with the M-channel ADCs (analog-to-digital converters) sampling at a fixed phase difference (e.g., 4 channels). The same signal is sampled at intervals. By using multiple low-speed ADCs in conjunction with a time-cross sampling mechanism, the sampling results of multiple low-speed ADCs are merged into an equivalent high-speed sampling data stream. At the same time, the ADC output data is received in DDR (double data rate) mode. Timing synchronization is achieved by using the FPGA's internal IDDR module and asynchronous FIFO. Ultimately, the Nyquist bandwidth is doubled to meet the high-speed acquisition requirements of pressure signals in collision tests.

[0079] Next, laser displacement sensors are deployed simultaneously to monitor seat belt slippage, tension sensors monitor webbing tension, and high-speed timers measure seat belt locking trigger time.

[0080] After the collision test is initiated, data such as pressure signal, slip, tension, and lock-up time are collected simultaneously. The raw data is stored in binary format, and the sampling time covers 50ms before the collision to 200ms after the collision to ensure complete capture of the entire collision process. At the same time, frequency domain analysis is used, and low-pass filtering is employed to separate the data. Steering wheel vibration and high-frequency seat belt load signal; combined with airbag pressure sensor data, a gas expansion dynamics model and finite element simulation were established. Utilizing the time difference between the seat belt and airbag triggering (15-30ms earlier for the seat belt), the forces exerted by the steering wheel and airbag on the seat belt were filtered out during the time-domain force differentiation phase, generating the final pressure acquisition results. Specifically:

[0081] Frequency domain analysis was used to separate the high-frequency seatbelt load signal through low-pass filtering. Specifically, after performing time-domain to frequency-domain signal conversion using Fourier transform, low-pass filtering is applied to separate the high-frequency seatbelt load signal. Finally, inverse Fourier transform is used to restore the frequency-domain to time-domain signal, yielding the restored time-domain high-frequency seatbelt load signal. .

[0082] A gas collision dynamics model and finite element simulation were established based on airbag pressure sensor data. The expression is as follows:

[0083]

[0084] in, This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. This represents the gas leakage correction factor, calibrated according to the airbag sealing test, with a value range of 0.92-0.98; express The mass of gas inside the airbag at any given time is obtained from the airbag gas-generating agent dosage and combustion rate model; This represents the gas constant, with a value of 296.8. ; express The gas temperature inside the airbag at any given time is calculated by coupling the gas combustion heat release model and the heat dissipation model. express The real-time volume of the airbag is measured, and the volume change curve of the airbag during the collision process is generated through finite element simulation.

[0085] The expression for the time difference between seatbelt and airbag deployment is:

[0086]

[0087] in, This indicates the time difference between the activation of the seat belt and the airbag. In engineering practice, the seat belt usually activates first, followed by the airbag. The airbag trigger time is indicated by actual measurement using an airbag pressure sensor. This indicates the seatbelt trigger moment, which is obtained from actual measurement by the seatbelt tension sensor.

[0088] During the time-domain force differentiation phase, the forces exerted by the steering wheel and airbag on the seatbelt are filtered out, generating the final pressure acquisition result, expressed as:

[0089]

[0090] in, This indicates that the final pressure data collection results only include the pure force exerted by the seatbelt on the dummy, and interference from the steering wheel and airbags has been eliminated. This represents the restored time-domain high-frequency seatbelt load signal. This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. This indicates the start time of the crash test, which is also the time reference point. Indicates the airbag trigger time. Indicates the end time of the crash test. This indicates the pure seatbelt operation phase, during which high-frequency seatbelt load signals are directly used. This indicates the stage of combined action of the seat belt and airbag, during which the airbag interference force needs to be eliminated.

[0091] S2: The pressure acquisition results are sequentially processed through signal amplification, signal noise identification and removal, data correction, and missing data supplementation to obtain the processed pressure acquisition results; wherein, S2 includes:

[0092] S2-1: The pressure acquisition results are synchronously amplified by an instrumentation amplifier and an operational amplifier, and then converted into a digital pressure signal;

[0093] S2-2: A second-order active low-pass filter is used to suppress 50Hz power frequency interference, and the pressure digital signal is sequentially processed by moving average filtering, Kalman filtering, and median filtering to obtain the preprocessed pressure digital signal.

[0094] S2-3: Real-time monitoring of the zero point of the flexible array pressure sensor, calculation of correction values ​​for the pressure digital signal based on a preset temperature-zero point drift third-order polynomial model, and elimination of abnormal drift points through a sliding window calculation.

[0095] S2-4: Real-time monitoring of the damaged pressure sensor at the damaged location in the flexible array pressure sensor, calling the undamaged pressure sensors around the damaged location for average value processing, and automatically filling the pressure data after processing with the pressure data of the damaged pressure sensor.

[0096] In this embodiment, the pressure analog signal in the acquired pressure data is synchronously amplified by a MAX1464 instrumentation amplifier and a non-inverting / differential operational amplifier circuit, and then processed using a 24-bit... The ADC converts the filtered original analog signal into a digital signal, then applies a weighted average filter to amplify the digital signal, and activates a second-order active low-pass filter to suppress 50Hz power frequency interference. Subsequently, a moving average filter is applied, followed by a Kalman filter, and a median filter is used to handle outliers. Dynamic noise covariance adjustment is also implemented.

[0097] Moving average filtering: Maintains a data queue of fixed length, removes the oldest data each time new data is added, and calculates the average of all data within the window as the output;

[0098] Kalman filtering describes the system dynamics through linear state equations and observation equations, and uses a recursive approach to achieve real-time state estimation.

[0099] Median filtering: A two-dimensional sliding template (such as 3×3, 5×5, or other odd-sized templates) is used to traverse the image, and the median value of the neighborhood gray values ​​of each pixel is taken after sorting them.

[0100] Finally, through real-time monitoring and dynamic correction, the zero-point drift problem in the sensor or measurement system is effectively eliminated. A third-order polynomial model of temperature-zero-point drift is established, and the coefficients are solved using the least squares method. This model is suitable for nonlinear drift scenarios. A radial basis function network is used to approximate complex drift characteristics, requiring fewer samples and achieving high fitting accuracy (error <0.1%). This is combined with sliding window statistics (such as...). Criteria) Remove outlier drift points (by calculating the mean of the data) ) and standard deviation ( ), will exceed Data within the specified range is considered outlier and removed, while valid signals are retained.

[0101] Finally, when local sensor damage is detected, data from surrounding sensors is automatically retrieved, the average value is calculated, and the data for the damaged area is filled in.

[0102] S3: Acquire raw image data of the dummy under the influence of the seatbelt during the crash test, construct a visualization interface of the dummy's stress surface contour, and calculate the average pressure value in the pressure acquisition results within a preset time window in real time as the current display threshold, and synchronously update the parameters of the dummy's stress surface contour visualization interface; wherein, S3 includes:

[0103] S3-1: Convert the pressure digital signal into pressure value, combine the physical position of the pressure sensor on the dummy with the mapping table of the unique identifier of the pressure sensor, assign the pressure value of each pressure sensor to the corresponding position of the virtual model of the visualization system, and use the 3D rendering engine to generate the contour map cloud of the dummy's force surface.

[0104] S3-2: Calculate the average pressure value in the pressure acquisition results within the preset time window in real time, use it as the current display threshold, and update the parameters of the dummy force surface contour visualization interface simultaneously.

[0105] In this embodiment, the digital signal is converted into a pressure value, and the pressure value of each pressure sensor is assigned to the corresponding position of the virtual model of the visualization system in combination with the ID-coordinate mapping table. The VTK 3D rendering engine is used to generate a dummy force surface contour cloud map with 10 levels of color gradient (blue 0-8kN, red 64-80kN).

[0106] Next, the mean or median of the data within the most recent time window is calculated as the current threshold to quickly respond to data fluctuations. This avoids the presence of too many interfering color blocks on the force surface outline in the visualization interface due to an excessively low threshold, and also avoids the force surface outline being unable to be displayed due to an excessively high threshold.

[0107] S4: Perform interpolation smoothing on the discrete pressure points of the pressure sensor in the visualization interface of the dummy's force surface contour; perform grayscale processing, image noise filtering, and image binarization on the original image data in the visualization interface of the dummy's force surface contour to obtain a binary image; wherein, S4 includes:

[0108] S4-1: In the visualization interface of the dummy's force surface contour, the discrete pressure points of the sampled pressure sensor are used to generate a smooth curve by optimizing the control points using a cubic B-spline interpolation algorithm to eliminate contour jaggedness.

[0109] S4-2: In the visualization interface of the dummy's stress surface contour, the original image data is converted into a grayscale image, Gaussian blur algorithm is applied to suppress noise, and adaptive thresholding method is used to convert the grayscale image into a binary image, fill the contour depression and eliminate burrs.

[0110] Among them, the sampled discrete pressure points are interpolated by B-spline, and the jagged edges are eliminated by control point optimization. The cubic B-spline basis function is used to ensure that the curve is second-order continuous and differentiable, thus meeting the smoothness requirements of the human body contour.

[0111] The original color image is converted to a grayscale image to simplify the processing dimensions. Then, Gaussian blur and other filtering algorithms are applied to suppress image noise and avoid the generation of false contours in subsequent steps. Image binarization is performed by setting a fixed threshold or using an adaptive thresholding method to convert the grayscale image into a binary image containing only pure black (background, pixel value 0) and pure white (foreground, pixel value 255). This makes the boundary of the target object clearly stand out, fills the contour depression, eliminates jagged edges and burrs, and forms a smooth line.

[0112] S5: The preset contour extraction algorithm is invoked to extract the core contour of the binary image and perform contour post-processing to obtain the final visualization image of the force contour of the seat belt-acting dummy; among which,

[0113] S5 includes:

[0114] S5-1: Call the OpenCV cv2.findContours() function, select the contour retrieval mode and contour approximation method, scan the preprocessed binarized image, output the contour point set and hierarchical information, and obtain the initial extracted dummy force surface contour.

[0115] S5-2: Call the OpenCV function cv2.contourArea() to calculate the contour area, and preset an area threshold to filter out small noise points. The expression is:

[0116]

[0117] The set of pixel coordinates of the contour is as follows: ;

[0118] A preset area threshold is established based on the minimum contact area of ​​the dummy's seat belt. ,when The outlines were identified as noise and removed.

[0119] S5-3: Real-time calculation of the contact area between the seat belt and the dummy, expressed as:

[0120]

[0121] in, for Rate of change of contact area at any time; for Actual contact area at any given moment; The initial contact area before the collision;

[0122] S5-4: Establish a pressure level-contact area change rate assessment table, based on real-time pressure values ​​and contact area change rates. Determine the risk level;

[0123] S5-5: Synchronously bind the dummy's stress surface contour cloud map with time series data, transmit it to the virtual model of the visualization system to establish a time axis, drag the time axis to display the stress contour back at any time, and mark and display the pressure data at the current time;

[0124] S5-6: Calculate the energy absorption distribution in the contact area of ​​the dummy, establish a histogram of regional energy proportion, calculate the energy concentration, and obtain the energy analysis results.

[0125] In this embodiment, the cv2.findContours() function in OpenCV is called to scan the preprocessed binary image. This function is based on an edge tracking algorithm to find all boundaries composed of consecutive white pixels. In this process, two key parameters need to be defined: one is the contour retrieval mode (such as RETR_EXTERNAL to retrieve only the outermost contour, or RETR_TREE to establish a complete contour hierarchy to handle nested objects), and the other is the contour approximation method (CHAIN_APPROX_SIMPLE to retain only the contour inflection point coordinates to compress the data, or CHAIN_APPROX_NONE to save all points on the boundary). The output of the function is a set of contour points and hierarchical information describing the topological relationship between contours.

[0126] Given that the initially extracted contours may contain a large number of irrelevant noise points, they need to be filtered according to the actual application requirements. Common criteria include geometric features such as contour area, perimeter, and circumscribed rectangle size. For example, cv2.contourArea() can be used to calculate the area and set a threshold to filter out small noise points. After filtering, the target contour can be drawn on the original image for visualization verification using the cv2.drawContours() function. Advanced geometric analysis of the contour can be performed, such as calculating its minimum circumscribed rectangle, minimum circumscribed circle, or performing polygon approximation, thereby providing a reliable data foundation for subsequent advanced vision tasks such as object measurement, shape recognition, and classification.

[0127] Next, the contact area change rate was calculated, and a two-dimensional evaluation table of pressure level and contact area change rate was established, as shown in Table 1 below:

[0128] Table 1. Two-dimensional evaluation table of pressure rating and contact area change rate.

[0129]

[0130] Among them, pressure level At 30, if the rate of change of contact area This indicates that the overall injury risk level for the corresponding part of the occupant is Level 1. The overall injury risk level is divided into 1-10 levels in this application, with Level 1 representing the lowest overall injury risk and Level 10 the highest. In this application, the overall injury risk level... Level 7 indicates a high risk, requiring optimization of the seatbelt structure or restraint parameters; as shown in Table 1 above, when the pressure level... At 50, the rate of change of contact area also At that time, the comprehensive injury risk level is 10, which means that there is a high probability of serious injury in the area, such as rib fracture or abdominal contusion, and the seat belt structure or restraint parameters need to be optimized.

[0131] Finally, the contour cloud map of the dummy's stress surface is synchronously bound with the time series data and transmitted to the virtual model of the visualization system.

[0132] The results of binding the dummy's stress surface contour cloud map with time series data are used to establish a time axis in the virtual model of the visualization system. By dragging the time axis, the stress contour at any time can be displayed back, and the pressure data at the current time can be marked and displayed.

[0133] Next, Radioss and HyperWorks software are integrated. HyperMesh is used for pre-processing of the collision scene, Radioss is used to calculate the collision energy equation, and HyperView is used for post-processing of the collision. The collision energy equation is the total energy of the system during the collision process, which is equal to the sum of kinetic energy, structural absorbed energy, dissipated energy and other energy components, and is the basis of collision energy analysis.

[0134] Subsequently, the energy absorption distribution in the dummy's contact area was calculated, a histogram of regional energy proportion was established, and the energy concentration was calculated to obtain the energy analysis results. Specifically, the energy absorption distribution in the dummy's contact area was determined by discretizing the dummy's seat belt contact area into finite element mesh elements, calculating the energy absorption and proportion of each area, and realizing the quantification of energy distribution.

[0135] Energy concentration reflects the energy absorption per unit area and is a core indicator for assessing the risk of localized damage. Its expression is:

[0136]

[0137] in, for Energy concentration in the target area at any given time. for The actual contact area of ​​region m at time m. This is the pressure gradient correction coefficient.

[0138] The energy analysis results are integrated with the force profile data. The force profile cloud map and energy distribution cloud map of the dummy are overlaid on the virtual model of the visualization system in the dummy force profile visualization interface and then visualized.

[0139] Finally, based on the above process steps, a multi-dimensional interpretation of the force profile is performed, including:

[0140] Static distribution analysis: Based on the contour cloud map of the dummy's stress surface, pressure concentration areas (such as the upper edge of the chest and the left side of the abdomen) are identified, and the average pressure, peak pressure, and pressure gradient of the concentration areas are calculated in combination with calibration data.

[0141] Dynamic evolution analysis: By backtracking the time series, we analyze the occurrence time of the pressure peak at the moment of impact (0-50ms), the location transfer trajectory (such as from the abdomen to the chest), and the impact of seat belt slippage on the force distribution.

[0142] Risk correlation analysis: By combining the injury risk table with energy concentration data, the likelihood of injury in high-risk areas is determined, and the seat belt locking time is correlated.

[0143] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A dynamic feature contour recognition method based on real-time measurement using a flexible array pressure sensor, characterized in that: include: S1: Flexible array pressure sensors and safety belts are sequentially deployed on the front of the dummy. During the crash test, the flexible array pressure sensors sample pressure data through a parallel hierarchical staggered sampling architecture, and the sampling results are merged through time cross-sampling to obtain the pressure acquisition results. S2: The pressure acquisition results are sequentially processed by signal amplification, signal noise identification and removal, data correction and missing data supplementation to obtain the processed pressure acquisition results; S3: Collect raw image data of the dummy under the action of the seat belt during the crash test, construct a visualization interface of the dummy's force surface contour, and calculate the average pressure value in the pressure acquisition results within the preset time window in real time as the current display threshold, and update the parameters of the dummy's force surface contour visualization interface synchronously. S4: Perform interpolation smoothing on the discrete pressure points of the pressure sensor in the visualization interface of the dummy's force surface contour; perform grayscale processing, image noise filtering, and image binarization on the original image data in the visualization interface of the dummy's force surface contour to obtain a binary image. S5: Call the preset contour extraction algorithm to extract the core contour of the binary image and perform contour post-processing to obtain the final visualization image of the force contour of the seat belt action dummy. Specifically, in S3, the original image data of the dummy under the action of the seat belt during the collision test is collected, and the visualization interface of the dummy's force surface contour is constructed by converting the pressure digital signal into pressure value, combining the physical position of the pressure sensor on the dummy and the mapping table of the unique identifier of the pressure sensor, assigning the pressure value of each pressure sensor to the corresponding position of the virtual model of the visualization system, and using a 3D rendering engine to generate the dummy's force surface contour map.

2. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 1, characterized in that: S1 includes: S1-1: Install a flexible array of pressure sensors on the front of the dummy, monitor the initial force information of the flexible array of pressure sensors, assign a unique identifier to each pressure sensor, and establish the correspondence between the pressure sensor and its physical position on the front of the dummy through a signal acquisition system. S1-2: Ensure the dummy is fully in contact with and secured to the seatbelt, keeping the dummy and seatbelt stationary; S1-3: Perform zero-point calibration, full-scale calibration, and linear calibration on the flexible array pressure sensor, and adjust the dummy and seat belt according to the calibration anomaly signal.

3. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 2, characterized in that: S1 further includes: S1-4: Construct a virtual model of the visualization system for seat belt effect dummy crash testing; S1-5: Conduct collision tests using a parallel hierarchical interleaved sampling architecture, including sampling the same signal through M ADCs with a fixed phase difference, with each ADC operating at a frequency of 1 / M of the total frequency; S1-6: The sampling results are merged through time cross-sampling, including controlling the phase difference of each clock through a programmable phase shifter circuit, receiving ADC output data in DDR mode, and performing timing synchronization with FPGA; S1-7: Synchronously monitor the seat belt slippage during the crash test, calculate the seat belt tension distribution unevenness, and monitor the seat belt locking trigger time; S1-8: The frequency domain analysis method is adopted to separate the high-frequency seat belt load signal through low-pass filtering; the gas expansion dynamics model and finite element simulation are established by combining the airbag pressure sensor data; by using the time difference between the seat belt and the airbag triggering, the force of the steering wheel and airbag on the seat belt is screened out in the time domain to distinguish the force action stage, and the final pressure acquisition result is generated.

4. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 3, characterized in that: Specifically, S1-8 are: Frequency domain analysis was used to separate the high-frequency seatbelt load signal through low-pass filtering. ; A gas collision dynamics model and finite element simulation were established based on airbag pressure sensor data. The expression is as follows: in, This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. This represents the gas leakage correction factor; express The mass of gas inside the airbag at all times. Represents the gas constant; express The temperature of the gas inside the airbag at all times. express Real-time volume of the airbag; The expression for the time difference between seatbelt and airbag deployment is: in, This indicates the time difference between the seatbelt and the airbag deployment. Indicates the airbag trigger time. Indicates the moment the seatbelt is triggered; During the time-domain force differentiation phase, the forces exerted by the steering wheel and airbag on the seatbelt are filtered out, generating the final pressure acquisition result, expressed as: in, This indicates the final pressure acquisition result. This represents the restored time-domain high-frequency seatbelt load signal. This represents the interference force of the airbag on the dummy calculated by the airbag gas collision dynamics model. Indicates the start time of the crash test. Indicates the airbag trigger time. Indicates the end time of the crash test. This indicates the stage where the seatbelt is in full effect. This indicates the combined action phase of the seatbelt and airbag.

5. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 3, characterized in that: S2 includes: S2-1: The pressure acquisition results are synchronously amplified by an instrumentation amplifier and an operational amplifier, and then converted into a digital pressure signal; S2-2: A second-order active low-pass filter is used to suppress 50Hz power frequency interference, and the pressure digital signal is sequentially processed by moving average filtering, Kalman filtering, and median filtering to obtain the preprocessed pressure digital signal. S2-3: Real-time monitoring of the zero point of the flexible array pressure sensor, calculation of correction values ​​for the pressure digital signal based on a preset temperature-zero point drift third-order polynomial model, and elimination of abnormal drift points through a sliding window calculation. S2-4: Real-time monitoring of the damaged pressure sensor at the damaged location in the flexible array pressure sensor, calling the undamaged pressure sensors around the damaged location for average value processing, and automatically filling the pressure data after processing with the pressure data of the damaged pressure sensor.

6. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 5, characterized in that: S4 includes: S4-1: In the visualization interface of the dummy's force surface contour, the discrete pressure points of the sampled pressure sensor are used to generate a smooth curve by optimizing the control points using a cubic B-spline interpolation algorithm to eliminate contour jaggedness. S4-2: In the visualization interface of the dummy's stress surface contour, the original image data is converted into a grayscale image, Gaussian blur algorithm is applied to suppress noise, and adaptive thresholding method is used to convert the grayscale image into a binary image, fill the contour depression and eliminate burrs.

7. The dynamic feature contour recognition method based on real-time measurement by a flexible array pressure sensor according to claim 6, characterized in that: S5 includes: S5-1: Call the OpenCV cv2.findContours() function, select the contour retrieval mode and contour approximation method, scan the preprocessed binarized image, output the contour point set and hierarchical information, and obtain the initial extracted dummy force surface contour. S5-2: Call the OpenCV function cv2.contourArea() to calculate the contour area, and preset an area threshold to filter out small noise points. The expression is: The set of pixel coordinates of the contour is as follows: ; A preset area threshold is established based on the minimum contact area of ​​the dummy's seat belt. ,when The outlines were identified as noise and removed. S5-3: Real-time calculation of the contact area between the seat belt and the dummy, expressed as: in, for Rate of change of contact area at any time; for Actual contact area at any given moment; The initial contact area before the collision; S5-4: Establish a pressure level-contact area change rate assessment table, based on real-time pressure values ​​and contact area change rates. Determine the risk level; S5-5: Synchronously bind the dummy's stress surface contour cloud map with time series data, transmit it to the virtual model of the visualization system to establish a time axis, drag the time axis to display the stress contour back at any time, and mark and display the pressure data at the current time; S5-6: Calculate the energy absorption distribution in the contact area of ​​the dummy, establish a histogram of regional energy proportion, calculate the energy concentration, and obtain the energy analysis results.

Images