An active detection system and method for quick-change battery fixation posture

Abstract

The application discloses a kind of active detection system and method of quick-change battery fixed situation, it is related to new energy vehicle battery swap safety monitoring field, active detection system is used to by mechanical locking and electrical connection through locking pre-tightening force physical coupling equivalent electromechanical network, including the characteristic signal generation and injection module, multidimensional feature acquisition module, feature comparison and situation cognition module, safety control and communication module connected in turn;Active detection method is applied to active detection system, including steps: active detection and signal acquisition;Feature extraction and health degree evaluation;Trend prediction and safety decision.The application regards mechanical locking and electrical connection as an equivalent electromechanical network coupled by locking pre-tightening force, uses the same excitation signal to excite synchronously, and realizes health quantitative evaluation by Mahalanobis distance analysis of electromechanical fusion feature vector, finally realizes the technical upgrade from "passive threshold alarm" to "active cognitive prediction" by trend prediction.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring for battery swapping in new energy vehicles, and in particular to an active detection system and method for a fixed battery swapping status. Background Technology

[0002] Battery swapping technology for electric vehicles is one of the effective ways to achieve rapid energy replenishment. The reliability of the swapped battery is ensured by mechanical locking, the swapping connector, and the cooling pipe interface. With the increase in the number of swaps, the locking mechanism inevitably experiences wear and increased gaps. Wear and fretting corrosion of the connector pin plating can lead to a non-linear increase in contact resistance, and the sealing performance of the cooling pipe may also degrade due to repeated insertion and removal. These problems are difficult to detect in the early stages using traditional displacement switches or temperature sensors. Once detected, they are often close to failure, which can easily lead to power interruption or thermal runaway risks during driving.

[0003] In the existing technology, there are solutions based on Hall sensors to detect the distance to the battery hanging point and temperature sensors to detect the terminal temperature. These solutions can only alarm when the distance exceeds the set value or react passively when the terminal heats up. There are also solutions that use locking mechanisms to monitor sensors to detect whether the locking hook is in place, but their detection dimension is limited to whether the physical position of the locking hook has been reached. There are also solutions that use dual-circuit contact to determine the continuity of the connector, but these can only perform switch quantity detection.

[0004] The common shortcomings of the above solutions are: relying solely on passive threshold judgments of a single physical quantity, they cannot capture early characteristic changes in the fixed mechanism under sub-health conditions, cannot comprehensively analyze the deterioration trend of mechanical and electrical coupling, and are all reactive responses lacking predictive capabilities. More importantly, the industry currently treats mechanical locking and electrical connection as two independent systems to be tested separately, failing to recognize that they are physically coupled through locking preload, thus making it impossible to obtain the overall normal state of electromechanical integration through synchronous detection. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned problems by providing an active detection system and method for fast-changing battery fixed status. The mechanical locking and electrical connection are considered as an equivalent electromechanical network physically coupled by the locking preload force. The same excitation signal is used for synchronous excitation, and health measurement is achieved through Mahalanobis distance analysis of electromechanical fusion feature vectors. Finally, trend prediction is used to achieve a technological upgrade from "passive threshold alarm" to "active cognitive prediction".

[0006] The technical solution adopted in this invention is as follows: An active detection system for a fast-changing battery in a fixed state, used to actively detect the equivalent electromechanical network formed by the mechanical locking and electrical connection of the fast-changing battery through the physical coupling of locking preload force, comprising a feature signal generation and injection module, a multi-dimensional feature acquisition module, a feature comparison and state recognition module, and a safety control and communication module connected in sequence; wherein: The feature signal generation and injection module is used to generate an excitation signal with spectral characteristics, and divide the excitation signal into two paths via a power divider: one path is injected into the mechanical locking of the equivalent electromechanical network via an exciter, and the other path is injected into the electrical connection of the equivalent electromechanical network via a directional coupler. A multi-dimensional feature acquisition module is used to simultaneously acquire the mechanical vibration response signal of the mechanical lock and the electrical transmission response signal of the electrical connection; The feature comparison and situational awareness module has a built-in benchmark feature model under normal conditions. The benchmark feature model includes a benchmark feature vector V_ref and a covariance matrix S. This feature comparison and situational awareness module is used to extract the real-time features of the mechanical vibration response signal and the electrical transmission response signal to construct a multi-dimensional real-time feature vector V(t), calculate the Mahalanobis distance D_M between V(t) and V_ref, and obtain the situational health index HI through a preset function mapping. The safety control and communication module is used to execute a graded control strategy for three states: normal, sub-healthy, and fault risk, based on the preset range into which the situational health index HI falls, and to interact with the vehicle controller and battery swapping facilities through communication.

[0007] Furthermore, the mechanical vibration response signal is acquired by a broadband vibration sensor integrated into the locking mechanism or battery pack housing; And / or the electrical transmission response signal is a reflected signal returned via the coupling port of the directional coupler, acquired by the amplitude and phase detection unit.

[0008] Furthermore, the real-time feature vector V(t) includes multiple spectral peak frequencies f and multiple half-power bandwidths BW extracted from the mechanical vibration response signal, as well as the normalized impedance real part Re(z) and imaginary part Im(z) at multiple preset frequency points extracted from the electrical transmission response signal.

[0009] Furthermore, the formula for calculating the Mahalanobis distance D_M is as follows: ; Furthermore, the baseline feature model also includes a classifier to distinguish between three states: normal, sub-healthy, and fault-risk, and output the corresponding posterior probabilities. The posterior probability of the normal state is denoted as P_n, the posterior probability of the sub-healthy state as P_s, and the posterior probability of the fault-risk state as P_f. The situational health index HI is obtained by weighted fusion of the health component HI_m based on Mahalanobis distance and the classifier health component HI_c. ; Wherein, HI_m is obtained from the Mahalanobis distance D_M via an exponential mapping function: ; k is the sensitivity calibration coefficient, which is determined through simulated loosening tests; ; c1 is the maximum value of the normal state range; c2 is the median value of the sub-healthy state range; c3 is the median value of the fault risk state range.

[0010] β is a weighting coefficient; if the baseline feature model does not have the classifier configured, then .

[0011] Furthermore, the security control and communication module divides the situation health index HI into the following intervals and performs hierarchical control: When HI ≥ the first threshold, it is determined to be in a normal state and a health message is sent. When the first threshold > HI ≥ the second threshold, it is determined to be a sub-healthy state, and a restricted operation request is sent. When HI < the second threshold, it is determined to be a fault risk state, and an emergency power-down command is sent.

[0012] Furthermore, the feature comparison and situational awareness module also has an adaptive algorithm built in. During the operation phase, when the situational health index HI is higher than the preset baseline update judgment value, the baseline update judgment value is greater than the lower limit of the preset interval in which the normal state is located. The baseline feature vector V_ref and covariance matrix S are dynamically updated with a forgetting factor to adapt to the baseline drift caused by institutional aging and environmental changes.

[0013] An active detection method for a fast-battery-changing fixed-state system, applied to the aforementioned active detection system, is characterized by comprising the following steps: S1: Active detection and signal acquisition; The feature signal generation and injection module periodically emits an excitation signal with spectral characteristics, and injects the excitation signal synchronously into the equivalent electromechanical network formed by the mechanical locking and electrical connection of the quick-change battery through the physical coupling of the locking preload force. The multi-dimensional feature acquisition module synchronously acquires the mechanical vibration response signal and electrical transmission response signal output by the equivalent electromechanical network. S2: Feature extraction and health assessment; The feature comparison and situational awareness module extracts real-time features of the mechanical vibration response signal and electrical transmission response signal to construct a real-time feature vector, calculates the Mahalanobis distance between the real-time feature vector and the baseline feature vector under normal conditions, and maps it to a situational health index. S3: Trend prediction and security decision-making; The security control and communication module performs linear regression or trend analysis on the time series of the situation health index to calculate the decay rate α; When the absolute value of the decay rate is greater than a preset threshold, and extrapolation prediction is performed based on the decay rate within a preset short-term prediction time window, if it is predicted that the index will fall below the first threshold or the second threshold, a graded control strategy is executed in advance.

[0014] Furthermore, the formula for calculating the attenuation rate α in step S3 is: ; Where t_i is the timestamp, HI_i is the status health index corresponding to the timestamp, and n is the number of data points.

[0015] Furthermore, when the decay rate α is less than 0 and its absolute value is greater than a preset degradation rate threshold, predictive control is performed: the remaining time to reach the first threshold is calculated, and if it is less than a preset warning time window, a warning is triggered; the remaining time to reach the second threshold is calculated, and if it is less than a preset forced intervention time window, the system is forced to enter limp mode.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention applies a closed loop of "active excitation-response acquisition-multidimensional feature comparison-health measurement" to the fixed status assessment of battery swapping, which can capture early slight loosening of the locking mechanism and small changes in electrical connection impedance, realizing a leap from "fault alarm" to "deterioration detection".

[0017] 2. This invention utilizes the long-neglected technical fact that mechanical locking and electrical connection are physically coupled through locking preload. It treats mechanical locking and electrical connection as equivalent electromechanical networks and uses the same excitation source for synchronous detection. In particular, it accurately models the correlation structure between mechanical and electrical features through the off-diagonal elements of the covariance matrix, and uses this structure to identify linkage degradation offset in Mahalanobis distance calculation. The resulting coupling degradation identification capability is an unpredictable technical effect that cannot be achieved by any independent detection or simple feature combination method.

[0018] 3. By extrapolating the trend of the situation health index time series, this invention can proactively trigger graded protection measures several minutes before the actual occurrence of a fault, reserving a sufficient time window, and completely changing the existing safety control mode of "passively responding only after a fault occurs".

[0019] 4. This invention uses a built-in adaptive algorithm to update the baseline feature model online with a forgetting factor, adaptively preventing baseline drift caused by normal wear and environmental temperature changes, and ensuring that the detection sensitivity does not degrade throughout the entire life cycle. Attached Figure Description

[0020] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the active detection system disclosed in this invention; Figure 2 This is a schematic diagram of the quick-swap battery system. Figure 3 This is a flowchart illustrating the active detection method disclosed in this invention; Figure 4 This is a schematic diagram of the baseline feature model update process disclosed in this invention; The diagram is labeled as follows: 1-Feature signal generation and injection module; 11-Power divider; 2-Equivalent electromechanical network; 21-Vibrator; 22-Directional coupler; 23-Mechanical lock; 24-Electrical connection; 3-Multi-dimensional feature acquisition module; 31-Vibration sensor; 32-Mechanical response acquisition unit; 33-Electrical response acquisition unit; 4-Feature comparison and situational awareness module; 41-Base feature model; 5-Safety control and communication module; 6-Battery assembly; 61-Battery pack body; 62-Quick-change connector plug; 63-Locking component; 7-Chassis; 71-Three-point positioning device; 72-Connector socket; 73-Vibration isolation element; 74-Chassis lock seat. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and through progressively illustrated embodiments.

[0022] Before elaborating on the detection system of this application, the target object to be detected will be described first.

[0023] See Figure 2 The quick-swap battery system includes a battery assembly 6 and a chassis 7. The battery assembly 6 includes a battery pack body 61, a quick-swap connector plug 62, and multiple locking components 63 distributed around the circumference of the battery pack body 61. The battery assembly 6 is connected to the chassis 7 through two core interfaces: a mechanical lock 23 and an electrical connection 24. The mechanical lock 23 and the electrical connection 24 are considered as an equivalent electromechanical network 2 physically coupled by a locking preload. Wherein: Electrical connection 24 is formed by the quick-change connector plug 62 and the connector socket 72 installed on the vehicle chassis 7. The quick-change connector plug 62 can be an integrated electrical connector and cooling pipe joint, or it can be just an electrical connector.

[0024] A chassis lock seat 74 is installed on the chassis 7 via a vibration isolation element 73. The locking element 63 cooperates with the chassis lock seat 74 to form a mechanical lock 23, which is used to securely lock the battery assembly 6 onto the chassis 7.

[0025] The chassis 7 is usually also equipped with a three-point positioning device 71, which includes a positioning pin and a guide groove, for positioning and guiding the battery pack body 61, ensuring that the battery pack body 61 can be accurately guided to the target position during installation.

[0026] The installation process of the battery pack 6 is designed strictly according to sequential logic: during installation, the battery pack body 61 is first lifted to the correct spatial position by the positioning guidance of the three-point positioning device 71; then the electrical connection 24 is completed, at which time the electrical connector and the cooling pipe sealing ring are simultaneously established to establish an initial connection; finally, the mechanical locking 23 is completed to firmly press the battery pack onto the chassis 7, completing the installation of the battery pack 6; the disassembly is carried out in the reverse order.

[0027] From the above structure and workflow, it is clear that the locking state of the mechanical lock 23 plays a crucial role in ensuring the stability of the electrical connection 24. When the locking mechanism becomes loose due to wear, the locking preload decreases. This change simultaneously affects the mechanical vibration transmission characteristics and the electrical contact impedance—that is, the mechanical lock 23 and the electrical connection 24 achieve physical coupling through the locking preload.

[0028] The mechanical locking 23 and the electrical connection 24 are physically coupled through a locking preload, as detailed below: The contact stiffness K of the mechanical lock 23 and the contact resistance Rc of the electrical connection 24 both vary with the locking preload F: the contact stiffness can be expressed as... Where n takes values ​​in the range [0.3, 0.8], and K0 is the reference stiffness under normal conditions; the contact resistance can be expressed as... Where m takes values ​​in the range [0.5, 1.0], and R_bulk is the inherent volume resistance of the electrical connector. When F decreases due to wear, the reduced contact stiffness causes the resonant peak frequency to shift downward, and the increased contact resistance causes the real part of the normalized impedance to rise.

[0029] It should be noted that R_bulk is introduced for more accurate modeling, illustrating that the resistance change of an electrical connector does not increase indefinitely; it has a physical lower limit (body resistance) and a degradation upper limit (contact resistance). When the locking preload F decreases, leading to poor contact, the main manifestations are... The sharp increase in the term, while R_bulk remains unchanged, allows the system to distinguish between "contact resistance degradation caused by mechanical loosening" and "basic impedance of electrical connectors," thereby improving the sensitivity of Mahalanobis distance identification of linkage offset.

[0030] It should be noted that the present invention considers the mechanical lock 23 and the electrical connection 24 as an equivalent electromechanical network 2. This network is located between the battery assembly 6 and the chassis 7 and is not a component of this active detection system. The excitation and acquisition of this active detection system are both applied to this equivalent electromechanical network 2.

[0031] The baseline characteristic model 41 is established under the condition that the battery assembly 6 is in a completely reliable fixed state (the fixed state is the normal state), and specifically includes the following steps: A1: Using a 30-second cycle, the excitation signal (10Hz-100kHz) generated by the feature signal generation and injection module 1 is synchronously injected into the mechanical lock 23 and the electrical connection 24 via the power divider 11. Upon receiving the excitation signal, the mechanical lock 23 generates a corresponding mechanical vibration response signal, and the electrical connection 24 generates a corresponding electrical transmission response signal. The multi-dimensional feature acquisition module 3 acquires the mechanical vibration response signal and the electrical transmission response signal, and extracts an M-dimensional feature vector V according to the method described in the following embodiment. If feasible, the dimension of the feature vector V is M=12. A2: Repeat step A1 to perform the measurement N times to obtain the health feature vector set. It is feasible; N=30. A3: Calculate the baseline eigenvector ; A4: Calculate the covariance matrix S, whose elements are: ; S_pq is the element in the p-th row and q-th column of the covariance matrix S.

[0032] S is an M×M symmetric matrix. The off-diagonal elements record the linear correlation between different features under normal conditions, especially the correlation structure characterizing the mechanical vibration response signal (such as resonant frequency) and the electrical transmission response signal (such as normalized impedance).

[0033] It should be noted that the robustness of the covariance matrix S has a fundamental impact on the accuracy of Mahalanobis distance calculation. To avoid the "masking effect" of outliers on covariance estimation, the present invention adopts the following measures in the establishment stage of the baseline feature model 41: First, all training samples are collected under a completely reliable fixed state confirmed by manual verification, and the installation status of the locking component 63 and the quick-change connector plug 62 is visually verified by the operator to be a completely reliable fixed state; Second, outlier detection is performed on the collected N sets of feature vectors (based on the temporary estimate of Mahalanobis distance, outlier samples with D_M>3 are removed) to ensure that the data participating in the modeling are typical samples under normal conditions. In the online operation stage, the update condition of the baseline model is set to the situation health index HI being continuously greater than the baseline update judgment value, and the baseline update judgment value being greater than the lower limit of the preset interval where the situation health index HI is located when the fixed situation is in a normal state. The forgetting factor λ is used to update V_ref and S by exponential weighted moving average, which further suppresses the influence of any potential outliers on the model; λ can be 0.95. The aforementioned engineering measures ensure that the covariance matrix S has good robustness throughout its entire life cycle, eliminating the need to introduce complex robust estimation algorithms such as MCD (minimum covariance determinant), thus reducing the computational burden on the vehicle controller.

[0034] Since a decrease in locking preload F simultaneously leads to a decrease in resonant frequency and an increase in the real part of contact impedance, the offset directions of these two characteristics during degradation (one decreasing and the other increasing, showing a negative correlation) are completely consistent with the statistical correlation direction generated by the small natural fluctuations of F under normal conditions—this correlation has been recorded in the off-diagonal elements of the covariance matrix S. Therefore, the directional characteristics of the linkage feature offset are inherently consistent with the correlation structure recorded in the covariance matrix S, which constitutes the physical basis for Mahalanobis distance's ability to sensitively identify such offsets.

[0035] During online monitoring, the deviation ΔV between the real-time feature vector V(t) and V_ref includes a linkage offset component caused by the decrease in locking preload, and its offset direction is consistent with the correlation direction defined by S under normal conditions. (Mahanobis distance) By assigning higher weight to the linkage direction using the inverse matrix of S, even minute linkage offsets can be detected sensitively; while non-linkage disturbances such as temperature drift and random vibrations are suppressed due to directional mismatch. This achieves highly sensitive and robust identification of electromechanical coupling degradation.

[0036] Example 1 See Figures 1 to 4 An active detection system for a fixed state of a quick-swap battery is used to actively detect the equivalent electromechanical network 2 formed by the mechanical locking 23 and electrical connection 24 of the quick-swap battery through the physical coupling of locking pre-tightening force. The active detection system includes a feature signal generation and injection module 1, a multi-dimensional feature acquisition module 3, a feature comparison and situation recognition module 4, and a safety control and communication module 5 connected in sequence.

[0037] The feature signal generation and injection module 1 can optionally use an AD9833 direct digital frequency synthesizer chip, which can be programmably used to generate any one of the following excitation signals with spectral characteristics: a swept frequency signal, a multi-frequency composite signal, or a pseudo-random sequence signal. Figure 2 As shown, the excitation signal generated by the feature signal generation and injection module 1 is split into two by the power divider 11 (the power divider 11 can be model ZFSC-2-6-S+): one path drives the exciter 21 (the exciter 21 can be a piezoelectric stack exciter 21 of model NAC2011) to generate micron-level amplitude vibration on the mechanical lock; the other path is injected into the electrical connection 24 through the directional coupler 22 (the directional coupler 22 can be a broadband directional coupler 22 of model ZFDC-10-5), thereby synchronously exciting the equivalent electromechanical network 2 formed by the mechanical lock 23 and the electrical connection 24.

[0038] The multi-dimensional feature acquisition module 3 is responsible for synchronously acquiring the response of the equivalent electromechanical network 2 to the excitation signal, and synchronously acquiring the mechanical vibration response signal of the mechanical lock 23 and the electrical transmission response signal of the electrical connection 24; the multi-dimensional feature acquisition module 3 includes a mechanical response acquisition unit 32 and an electrical response acquisition unit 33; wherein: The mechanical response acquisition unit 32 includes a signal conditioning circuit and an analog-to-digital converter (ADC). The mechanical vibration response signal is picked up by the vibration sensor 31 (such as a front-end broadband vibration sensor 31 of model ADXL1002) and converted into an analog signal. After anti-aliasing filtering and amplitude adjustment by the signal conditioning circuit, it is converted into a digital vibration signal by the ADC at a sampling rate of 200 kSps for processing by the feature comparison and situational awareness module 4. In this embodiment, the signal conditioning circuit can use an ADA4622 operational amplifier to form a second-order active low-pass filter, and the ADC can use an AD7606 type 8-channel synchronous sampling ADC.

[0039] The complete transmission chain for the mechanical lock 23 to be excited and generate a mechanical vibration response is as follows: exciter 21 → chassis lock seat 74 → locking mating surface → locking component 63 → battery pack body 61 → vibration sensor 31 → mechanical response acquisition unit 32. The exciter 21 is fastened to the lock seat base and converts the excitation electrical signal into mechanical vibration, which directly acts on the contact mating surface between the lock hook and the lock seat. The vibration sensor 31 is rigidly attached to the battery pack body 61 near the mating surface, picking up the vibration response transmitted through the above path and converting it into an analog electrical signal. When the locking mechanism experiences increased gaps due to wear, the contact stiffness of the mating surface decreases, and the vibration transmission characteristics of the above transmission chain change accordingly, manifested as a shift in the peak frequency and an increase in the half-power bandwidth, thus reflecting a loose state.

[0040] It should be noted that the excitation vibration generated by the exciter 21 is a micro-vibration with an amplitude in the micrometer range, and its energy is concentrated in the high-frequency band above 1kHz. Its effect is limited to a local area of ​​the locking mechanism's mating surface. Since high-frequency vibrations decay extremely quickly in metal structures, and there are vibration isolation elements 73 between the lock seat and the vehicle chassis 7, this micro-vibration cannot be transmitted to the passenger compartment, and its impact on the vehicle's ride comfort is minimal. As a further safety guarantee, the system can also optionally be equipped with a vehicle vibration acceleration safety monitoring mechanism. When this optional mechanism is activated, if abnormal vibration propagation is detected, the excitation will be immediately stopped to ensure that active detection does not affect the vehicle's original NVH performance.

[0041] Furthermore, the system's operating mode can be optimized according to the vehicle's status: In parking and battery swapping mode, when the vehicle is stationary and unoccupied, the system can perform full-band frequency sweep excitation to obtain the most complete electromechanical response characteristics; in driving mode, the system adopts an intermittent working mode, exciting for a specific duration at set intervals, such as only 0.5 seconds every 30 seconds, with a duty cycle of approximately 1 / 50, ensuring timely monitoring while minimizing excitation energy. Within this 0.5-second excitation window, 100k data points can be collected at a sampling rate of 200kSps, sufficient to perform multiple 4096-point FFTs and take the average, with a frequency resolution of approximately 48.8Hz, which can clearly distinguish the shift in the characteristic frequency of the locking mechanism. The 30-second monitoring interval balances the timely detection of gradual degradation with reasonable control of system power consumption. For extreme situations such as sudden breakage, this active detection system also incorporates the real-time situational health index (HI) threshold for fallback protection. The two modes work together to ensure the continuity and accuracy of situational awareness while eliminating any potential impacts during driving.

[0042] The electrical response acquisition unit 33 can be an AD8302 type amplitude and phase detection unit to receive the reflected signal output from the coupling end of the directional coupler 22. Through the internal logarithmic amplifier and phase detector, it outputs the amplitude ratio VMAG and phase difference VPHS between the reflected signal and the original excitation signal for subsequent vector network parameter calculation and real-time feature extraction.

[0043] The feature comparison and situational awareness module 4 can use an automotive-grade microcontroller, model SAK-TC397XP-256F300S, to perform feature extraction, comparison, and awareness. (1) Extract real-time features and construct a multi-dimensional real-time feature vector: The mechanical vibration signal was subjected to a 50kHz low-pass filter and a 4096-point Fast Fourier Transform (FFT) to obtain the spectrum. The center frequencies f1, f2, and f3 corresponding to the three largest resonance peaks were extracted, along with their half-power bandwidths BW1, BW2, and BW3 at the -3dB amplitude.

[0044] For the electrical transmission response signal, the amplitude and phase of the reflection coefficient S11 are calculated based on the VMAG and VPHS output by the electrical response acquisition unit 33: , , The calculated reflection coefficient is: .

[0045] The reflection coefficient S11 is a vector network parameter that fully describes the signal reflection characteristics of the electrical connector at the current excitation frequency.

[0046] To eliminate the influence of transmission line characteristic impedance and make measurement results at different frequencies comparable, the relationship between reflection coefficient and normalized impedance in transmission line theory is used. Convert S11 to the real part Re(z) and imaginary part Im(z) of the normalized impedance: ; .

[0047] in Z0 is the characteristic impedance of the transmission line (50Ω). When the quick-connect connector has good contact and impedance matching, When there is poor contact or impedance mismatch, z deviates significantly from the ideal value.

[0048] In subsequent feature extraction, multiple preset frequency points such as 1kHz, 10kHz and 80kHz are selected, and the normalized impedance real part Re(z) and imaginary part Im(z) are calculated according to the above steps, respectively, as real-time features of the electrical transmission response signal.

[0049] This constitutes an M-dimensional (as described above, 12-dimensional) real-time feature vector: .

[0050] (2) Health assessment; see Figure 3 Calculate the Mahalanobis distance D_M and map the health component HI_m based on the Mahalanobis distance; where: Mahalanobis distance D_M: ; Based on the health component HI_m of Mahalanobis distance, 0 < HI_m ≤ 100: ; Where k is the sensitivity calibration coefficient, determined through simulated loosening tests. In this embodiment, a k value of 0.05 yields good discrimination.

[0051] To further enhance the robustness of the assessment, a classifier can be optionally added to the baseline model. This classifier uses a linear support vector machine, trained on labeled samples (normal, sub-healthy, and fault risk) collected in a simulated loosening test. It distinguishes between the three states of normal, sub-healthy, and fault risk and outputs the corresponding posterior probabilities. During online operation, the classifier outputs three posterior probabilities for the current real-time feature vector V(t), where the posterior probability of the normal state is denoted as P_n, the posterior probability of the sub-healthy state is denoted as P_s, and the posterior probability of the fault risk state is denoted as P_f. Based on the posterior probabilities P_n, P_s, and P_f, the health component HI_c of the classifier is calculated, where 0 < HI_c ≤ 100. .

[0052] Where c1 is the maximum value of the normal state interval; c2 is the median value of the sub-healthy state interval; and c3 is the median value of the fault risk state interval. If the first threshold is 80 and the second threshold is 60, then... , .

[0053] Finally, the situational health index HI is a weighted fusion of the health component HI_m based on Mahalanobis distance and the classifier health component HI_c: .

[0054] β is a weighting coefficient; if the baseline feature model 41 is not configured with the classifier, then Of course, in this embodiment, β=0.45. This scheme combines the advantages of statistical distance and pattern recognition, and is a further limitation rather than a replacement for the main scheme.

[0055] After obtaining the situational health index HI through the above weighted fusion calculation, the safety control and communication module 5 is used to execute a graded control strategy of three levels—normal, sub-healthy, and fault risk—based on the preset range in which the situational health index HI falls, and interacts with the vehicle controller and battery swapping facilities through communication. The specifics are as follows.

[0056] When HI≥ the first threshold, the first threshold can be equal to 80, which is considered a normal state. Normal state messages are periodically sent to the vehicle controller via the CAN bus, and the safety control and communication module 5 drives the status indicator light to show green, indicating that the vehicle can operate normally at full power. When the first threshold > HI ≥ the second threshold, the second threshold can be equal to 60, which is judged as a sub-healthy state. A restricted operation message is sent to the vehicle controller via the CAN bus to request that the vehicle speed be limited to within 40km / h and the limp mode be activated. At the same time, the safety control and communication module 5 drives the indicator light to turn yellow, and the instrument panel displays the warning information simultaneously. When HI < the second threshold, it is determined to be a fault risk state. An emergency power-down message is sent to the vehicle controller via the CAN bus. At the same time, the safety control and communication module 5 drives the indicator light 51 to turn red, and the instrument panel displays the warning information synchronously.

[0057] The basis for setting the hierarchical control threshold is: In this embodiment, the first threshold is preferably 80, and the second threshold is preferably 60. The above-mentioned grading thresholds are determined by a combination of simulated loosening accelerated test and statistical analysis: under ideal fastening conditions, the situational health index HI value is concentrated in the high range of 95-100. After successively introducing simulated loosening, the situational health index HI value shows a monotonically decreasing trend. The first threshold of 80 is based on the range where the Health Index (HI) value begins to show a significant separation from the normal state, ensuring that it has exceeded the normal tolerance fluctuation range. The second threshold of 60 is set at the starting range of the situation health index (HI) value showing an accelerated downward trend, leaving sufficient safety margin.

[0058] Using this active detection system, the battery pack 6 (384kg power battery, using 6 locking mechanisms) of a new energy vehicle was gradually changed from a state of being firmly locked to the chassis 7 to a state of being disconnected. Simultaneously, the locking preload was detected using a force measuring device (the force measuring device uses a Kistler 9331B piezoelectric force sensor; this device and detection method are well-known in the industry). The detection performance of the active detection system was tested based on the locking preload. The detected state and the actual state are shown in Table 1 below. Table 1: Comparison of Fixed-Situation Detection Results with Actual Results

[0059] As can be seen from the data in Table 1, the active detection system disclosed in this embodiment has a good detection effect. This further illustrates that the benchmark feature model established by the method disclosed in this invention has a better benchmark reference effect in the active detection system.

[0060] One possible implementation method is described in [reference needed]. Figure 4 The feature comparison and situational awareness module 4 also has a built-in adaptive algorithm. During the operation phase, when the situational health index HI is higher than the preset baseline update judgment value, the baseline update judgment value is greater than the lower limit of the preset interval in which the normal state is located. For example, the baseline update judgment value is set to 85, which is slightly higher than the first threshold. The baseline feature vector V_ref and covariance matrix S are dynamically updated with a forgetting factor to adapt to the baseline drift caused by institutional aging and environmental changes. The specifics are as follows.

[0061] When the situation health index HI remains higher than the baseline update judgment value (e.g., the baseline update judgment value is set to 85), V_ref and S are updated using an exponentially weighted moving average with a forgetting factor λ: , , Wherein, V_ref_new is the updated baseline feature vector; V_ref_old is the baseline feature vector before the update; and V_new is the real-time feature vector in the latest normal state. S_new is the updated covariance matrix; S_old is the covariance matrix before the update; .

[0062] In this embodiment, the forgetting factor λ can be set to 0.95, with an equivalent time constant of approximately 20 update cycles. This effectively smooths out short-term random fluctuations while sensitively tracking slow baseline drift caused by normal wear or seasonal temperature changes. The forgetting factor can also be adjusted within the range of 0.9 to 0.99 based on the actual vehicle model's battery swapping frequency.

[0063] The baseline update decision value is set to 85, which is near the lower limit of the normal state range. The current sample is only included in the update when the situation health index (HI) remains above this value, ensuring that degraded data does not contaminate the baseline model. The key to this design is that the baseline update decision value of 85 is higher than the lower limit of the normal state (80). Therefore, even in the early sub-healthy stage (e.g., HI=75), the system will lock the baseline model to prevent data that has deviated from the normal state from erroneously lowering the baseline, thereby ensuring the purity and sensitivity of the model throughout its entire lifecycle.

[0064] It should be noted that the benchmark update judgment value has a different function from the first threshold and the second threshold mentioned above: the first threshold and the second threshold are used for safety control decisions, while the benchmark update judgment value is only used for model adaptive update judgment.

[0065] In addition to the aforementioned real-time threshold control, this active detection system also possesses predictive control capabilities: when the trend prediction indicates that the situation health index HI will fall below the corresponding threshold within a preset time window, the corresponding control action is triggered before the situation health index HI actually falls below the threshold. Compared with passive real-time threshold control, the necessity of predictive control lies in its ability to identify irreversible deterioration trends when the equivalent electromechanical network 2 is still in a sub-healthy or even normal range, thereby taking power reduction or limp-out measures several minutes in advance, reserving valuable time for the driver to find a safe stopping point, and effectively solving the problem of delayed response of passive alarms in the face of sudden fault precursors. This predictive control logic and real-time threshold control constitute a two-layer progressive protection mechanism. For details of the prediction method and triggering logic, please refer to step S3 disclosed in Embodiment 2 below.

[0066] In this embodiment, the control logic and CAN communication protocol stack of the safety control and communication module 5 run on the independent core of the SAK-TC397XP microcontroller, sharing the same hardware platform with the feature comparison and situation awareness module 4. Functional isolation is achieved through software partitioning to meet functional safety requirements.

[0067] In addition, the safety control and communication module 5 can also integrate a battery swapping mode safety interlock function: before battery swapping, it verifies three conditions: vehicle speed is zero, parking gear / neutral, and parking brake is active. Only after these conditions are met can the battery swapping mode be entered, and the module takes over control of the vehicle controller and locking motor. After battery swapping, the safety control and communication module 5 returns control, and the feature comparison and situation awareness module 4 then reads the battery pack's unique identification code via the CAN bus: if the ID remains unchanged and the self-check situation health index HI recovers to above 85, the original baseline feature model 41 is used; if the ID changes, the baseline feature model 41 corresponding to the battery pack is downloaded from the vehicle storage or cloud. If it cannot be obtained, the initialization program is automatically executed to re-establish the baseline. This mechanism ensures that the baseline model accurately matches the actual normal state of the current battery pack after each battery swap.

[0068] Example 2 This invention also provides an active detection method for a fast-changing battery in a fixed state, applied to the active detection system described in Example 1. The overall process is described below. Figure 3 This includes the following steps: S1: Active Detection and Signal Acquisition; During vehicle operation or battery swapping, the feature signal generation and injection module 1 emits an excitation signal with spectral characteristics (sweep range 10Hz-100kHz, peak-to-peak value 5V) with a 30-second cycle. This excitation signal is divided into two paths by the power divider 11: one path drives the piezoelectric stack exciter 21, which acts on the locking mechanism's mating surface to generate micron-level mechanical vibration; the other path is injected into the detection circuit of the quick-change connector via the broadband directional coupler 22. The multi-dimensional feature acquisition module 3 acquires the mechanical vibration response signal through the mechanical response acquisition unit 32 (200kSps sampling rate) and the electrical transmission response signal through the electrical response acquisition unit 33. The above two signals are sent to the feature comparison and situational awareness module 4 after analog-to-digital conversion. In driving mode, the active detection system is excited for only 0.5 seconds every 30 seconds (duty cycle 1 / 50). Within the 0.5-second window, 100k sampling points can support multiple 4096-point FFT averaging, with a frequency resolution of about 48.8Hz, which is sufficient to distinguish the shift of the characteristic frequency of the locking mechanism, while ensuring that there is no substantial impact on ride comfort. In parking and battery swapping mode, it can perform continuous frequency sweep across the entire frequency band.

[0069] S2: Feature extraction and health assessment; Feature comparison and situational awareness module 4 processes the acquired signals: The mechanical vibration signal is low-pass filtered at 50kHz and then subjected to a 4096-point FFT to extract the first three resonant frequencies f1, f2, and f3, as well as the half-power bandwidths BW1, BW2, and BW3. The electrical signal is processed by S11 reflection coefficient calculation and normalized impedance transformation to extract Re(z) and Im(z) at three frequency points: 1kHz, 10kHz, and 80kHz, thus forming a 12-dimensional real-time feature vector V(t). The feature comparison and situation awareness module 4 reads the baseline feature vector V_ref and covariance matrix S from the memory, calculates the Mahalanobis distance, and finally maps it to a situation health index of 0-100. This situation health index HI value comprehensively reflects the overall reliability of the mechanical lock 23 and the electrical connection 24: 100 indicates complete consistency with the normal state, and 0 indicates serious deviation.

[0070] S3: Trend prediction and security decision-making; Feature comparison and situational awareness module 4 maintains a situational health index HI time series buffer of length n=20, covering 10 minutes, and updating every 30 seconds; The least squares method is used to perform linear fitting on the most recent n=20 situational health index HI data points to obtain the decay rate α: .

[0071] Where t_i is the timestamp, HI_i is the status health index corresponding to the timestamp, and n is the number of data points.

[0072] Accelerated degradation tests verified that during the linear decrease of locking preload, the Health Index (HI) drops from 80 to the sub-healthy range of 60, exhibiting a highly linear decay trend. The coefficient of determination (R²) obtained using the least squares linear regression fitting described above is greater than 0.95. Therefore, within the set short-term prediction time window (5 minutes in this embodiment), linear extrapolation can provide accurate and reliable trend predictions. By strictly limiting the prediction extrapolation to this short-term window, the system effectively avoids prediction errors that may be introduced by long-term nonlinear factors.

[0073] When α < 0 and |α| > 0.05 (unit: health index / minute), it indicates a continuous deterioration trend, and prediction is initiated. The general formula for predicting the remaining time to reach the first and second thresholds is: ; ; Where: HI_current is the current situation health index; t_rem1 is the remaining time to reach the first threshold; t_rem2 is the remaining time to reach the second threshold.

[0074] A yellow light warning is triggered when t_rem1 < 5 minutes; limp mode is forcibly entered when t_rem2 < 2 minutes.

[0075] The first threshold and the second threshold are the corresponding thresholds defined in Example 1, and in this example, they are specifically set to 80 and 60.

[0076] This predictive control, together with real-time threshold control, constitutes a two-layer progressive protection: Predictive control: If the unequal situation health index HI actually falls below the threshold, control actions are triggered in advance to buy time for safe handling. When t_rem1 < 5 minutes, a yellow light warning is triggered, with the warning time point being... t_current represents the current time point; when t_rem2 < 2 minutes, limp mode is forcibly entered, limiting the vehicle speed to below 40 km / h, allowing sufficient time for the driver to safely stop. The value of predictive control lies in early triggering. For example, if the situation health index HI is predicted to fall below 60, proactive limp mode can be initiated. At this time, the situation health index HI may still be within the normal range (≥80) or the sub-healthy range (80-60), which provides more time for response than passive limp mode after the situation health index HI actually falls below 60.

[0077] Real-time threshold control: If predictive control is not triggered (e.g., a sudden deterioration causes the situation health index HI to drop instantly), or if the situation health index HI continues to deteriorate and actually falls below the threshold during the execution of limp mode, the system retains real-time threshold triggering as a fallback protection: when the situation health index HI actually falls below 80, it will forcibly enter limp mode; when the situation health index HI actually falls below 60, it will execute an emergency power-down command. This is the last line of defense.

[0078] The criteria for setting the tiered early warning time window are as follows: 5-minute warning window: Taking into account three time components—driver reaction time 0.5-1 minute, time required for the vehicle to reach the nearest safe stopping point at a limp speed of 40km / h 2-4 minutes, and safety margin for dealing with congestion and other unexpected events 0.5-1 minute, the three components combined are about 3-6 minutes, ensuring that the driver has enough time to complete the parking, taking a conservative value of 5 minutes. Two-minute mandatory intervention window: According to the simulated accelerated degradation test, the situation health index HI value drops from 60 to the critical value of structural detachment HI≈30 at a typical rate of 0.05-0.1 / min in 2-3 minutes. Two minutes is taken as the mandatory intervention window to ensure that the high voltage is cut off before failure.

[0079] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. An active detection system for a fast-changing battery in a fixed state, used for actively detecting the equivalent electromechanical network (2) formed by the mechanical locking (23) and electrical connection (24) of the fast-changing battery through the physical coupling of locking preload, characterized in that: It includes a feature signal generation and injection module (1), a multi-dimensional feature acquisition module (3), a feature comparison and situation awareness module (4), and a security control and communication module (5), which are connected in sequence; wherein: The feature signal generation and injection module (1) is used to generate an excitation signal with spectral characteristics and divide the excitation signal into two paths via a power divider (11): one path is injected into the mechanical lock (23) of the equivalent electromechanical network (2) via an exciter (21), and the other path is injected into the electrical connection (24) of the equivalent electromechanical network (2) via a directional coupler (22). The multi-dimensional feature acquisition module (3) is used to simultaneously acquire the mechanical vibration response signal of the mechanical lock (23) and the electrical transmission response signal of the electrical connection (24); The feature comparison and situation cognition module (4) has a built-in benchmark feature model (41) under normal conditions. The benchmark feature model (41) includes a benchmark feature vector V_ref and a covariance matrix S. The feature comparison and situation cognition module (4) is used to extract the real-time features of the mechanical vibration response signal and the electrical transmission response signal to construct a multi-dimensional real-time feature vector V(t), calculate the Mahalanobis distance D_M between V(t) and V_ref, and obtain the situation health index HI through a preset function mapping. The safety control and communication module (5) is used to execute a graded control strategy for three states: normal, sub-healthy, and fault risk, based on the preset range into which the situation health index HI falls, and to interact with the vehicle controller and battery swapping facilities through communication.

2. The active detection system according to claim 1, characterized in that: The mechanical vibration response signal is collected by a broadband vibration sensor (31) integrated on the locking mechanism or battery pack housing; And / or the electrical transmission response signal is a reflected signal returned via the coupling port of the directional coupler (22), acquired by the amplitude and phase detection unit.

3. The active detection system according to claim 1, characterized in that: The real-time feature vector V(t) includes multiple peak frequencies f and multiple half-power bandwidths BW extracted from the mechanical vibration response signal, as well as the real part Re(z) and imaginary part Im(z) of the normalized impedance at multiple preset frequency points extracted from the electrical transmission response signal.

4. The active detection system according to claim 1, characterized in that: The formula for calculating the Mahalanobis distance D_M is: 。 5. The active detection system according to claim 1, characterized in that: The baseline feature model (41) also includes a classifier to distinguish between three states: normal, sub-healthy, and fault risk, and output the corresponding posterior probabilities. The posterior probability of the normal state is denoted as P_n, the posterior probability of the sub-healthy state as P_s, and the posterior probability of the fault risk state as P_f. The situational health index HI is obtained by weighted fusion of the health component HI_m based on Mahalanobis distance and the classifier health component HI_c. ； Wherein, HI_m is obtained from the Mahalanobis distance D_M via an exponential mapping function: ； k is the sensitivity calibration coefficient, which is determined through simulated loosening tests; ； c1 is the maximum value of the normal state interval; c2 is the median value of the sub-healthy state interval; c3 is the median value of the fault risk state interval. β is the weighting coefficient; if the baseline feature model (41) does not have the classifier configured, then .

6. The active detection system according to claim 1, characterized in that: The security control and communication module (5) divides the situation health index HI into the following intervals and performs hierarchical control: When HI ≥ the first threshold, it is determined to be in a normal state and a health message is sent. When the first threshold > HI ≥ the second threshold, it is determined to be a sub-healthy state, and a restricted operation request is sent. When HI < the second threshold, it is determined to be a fault risk state, and an emergency power-down command is sent.

7. The active detection system according to claim 1, characterized in that: The feature comparison and situation cognition module (4) also has an adaptive algorithm built in; when the situation health index HI is higher than the preset baseline update judgment value during the operation phase, the baseline update judgment value is greater than the lower limit of the preset interval where the normal state is located, and the baseline feature vector V_ref and covariance matrix S are dynamically updated with forgetting factor to adapt to the baseline drift caused by mechanism aging and environmental changes.

8. An active detection method for a fixed battery swapping configuration, applied to the active detection system described in any one of claims 1-7, characterized in that: S1: Active detection and signal acquisition; The feature signal generation and injection module (1) periodically emits an excitation signal with spectral characteristics, and injects the excitation signal synchronously into the equivalent electromechanical network (2) formed by the mechanical locking (23) and electrical connection (24) of the quick-change battery through the power divider (11) via the locking preload physical coupling. The multi-dimensional feature acquisition module (3) synchronously acquires the mechanical vibration response signal and electrical transmission response signal output by the equivalent electromechanical network (2). S2: Feature extraction and health assessment; The feature comparison and situational awareness module (4) extracts the real-time features of the mechanical vibration response signal and the electrical transmission response signal to construct a real-time feature vector, and calculates the Mahalanobis distance between the real-time feature vector and the benchmark feature vector under normal conditions, and maps it to the situational health index; S3: Trend prediction and safety decision-making; The safety control and communication module (5) performs linear regression or trend analysis on the time series of the situation health index and calculates the decay rate α; When the absolute value of the decay rate is greater than the preset threshold, and extrapolation prediction is performed based on the decay rate within the preset short-term prediction time window, and it is predicted that the first threshold or the second threshold will be broken, the hierarchical control strategy is executed in advance.

9. The active detection method according to claim 8, characterized in that: The formula for calculating the attenuation rate α in step S3 is: ； Where t_i is the timestamp, HI_i is the situation health index corresponding to the timestamp, and n is the number of data points.

10. The active detection method according to claim 8, characterized in that: When the decay rate α is less than 0 and its absolute value is greater than the preset degradation rate threshold, predictive control is performed: calculate the remaining time to reach the first threshold, and if it is less than the preset warning time window, trigger a warning. Calculate the remaining time to reach the second threshold. If it is less than the preset mandatory intervention time window, force entry into limp mode.

Images