Flatness detection method of power battery cold plate and related equipment

Abstract

The invention discloses a flatness detection method of a power battery cold plate and related equipment, and relates to the technical field of power batteries, the method comprises the following steps: when a plurality of ultrasonic sensors transmit ultrasonic pulse signals to the surface of the power battery cold plate, obtaining reflection signals reflected by the surface of the power battery cold plate and received by a target ultrasonic sensor, wherein the plurality of ultrasonic sensors comprise a target ultrasonic sensor; generating a frequency domain characteristic spectrum of the target ultrasonic sensor based on the reflected signal; inputting the frequency domain characteristic spectrum into a fusion characteristic network to generate a fusion characteristic spectrum for representing the flatness of the power battery cold plate; and inputting the fused characteristic spectrum into a flatness identification model, and outputting a flatness detection result of the power battery cold plate. According to the invention, accurate detection of the flatness of the power battery cold plate is realized through multi-sensor cooperative acquisition, frequency domain feature fusion and an intelligent identification model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power batteries, in particular to a flatness detection method of a power battery cold plate and related equipment. BACKGROUND

[0002] The working temperature of a power battery directly affects its performance, safety and cycle life, and the flatness quality of the power battery cold plate as a core component of the thermal management system is crucial to the assembly accuracy, structural stability and heat conduction efficiency of the battery pack. If the flatness of the cold plate is insufficient, it will reduce the contact area with the power battery module and reduce the heat conduction performance; at the same time, it may cause uneven flow of the cooling liquid, causing local overheating or overcooling of the battery temperature field, which seriously affects the safety of the battery.

[0003] In the prior art, a three-coordinate measuring instrument is usually used to detect the flatness of the power battery cold plate. However, this method has obvious shortcomings: on the one hand, the three-coordinate measuring instrument has high equipment purchase cost and needs to be regularly maintained and calibrated by professional technicians, which is costly; on the other hand, it uses a point-by-point data collection method, which requires clamping, positioning and calibration of the workpiece before measurement, resulting in long measurement time and low efficiency, which is difficult to meet the detection needs of mass production. Therefore, there is an urgent need for a flatness detection method of a power battery cold plate to solve the above-mentioned technical problems. SUMMARY

[0004] A series of simplified concepts are introduced in the summary section, which will be further described in detail in the specific embodiments section. The summary section of the present application does not mean to attempt to limit the key features and necessary technical features of the claimed technical solution, nor does it attempt to determine the protection scope of the claimed technical solution.

[0005] The present application aims to solve the problems of high cost, long measurement time and low efficiency of the flatness detection of the power battery cold plate in the prior art, and realizes efficient and accurate detection of the flatness of the power battery cold plate through multi-sensor collaborative collection, frequency domain feature fusion and intelligent recognition model.

[0006] In a first aspect, the present application provides a flatness detection method of a power battery cold plate, comprising:

[0007] When a plurality of ultrasonic sensors emit ultrasonic pulse signals to the surface of the power battery cold plate, a reflected signal reflected by the surface of the power battery cold plate and received by a target ultrasonic sensor is acquired, wherein the plurality of ultrasonic sensors includes the target ultrasonic sensor;

[0008] Based on the reflected signal, a frequency domain feature spectrum of the target ultrasonic sensor is generated;

[0009] The frequency domain feature map is input into a fusion feature network to generate a fusion feature map used for representing the flatness of the power battery cooling plate.

[0010] The fusion feature map is input into a flatness recognition model to output a flatness detection result of the power battery cooling plate.

[0011] In some embodiments, the spatial layout of the plurality of ultrasonic sensors satisfies that the vertical projection points of the plurality of ultrasonic sensors on the surface of the power battery cooling plate are in a continuous polyline distribution.

[0012] In some embodiments, based on the reflection signal, a frequency domain feature map of the target ultrasonic sensor is generated, including:

[0013] An analog-to-digital conversion operation is performed on the reflection signal to generate a digital signal;

[0014] A Fourier transform is performed on the digital signal to generate an initial frequency spectrum;

[0015] Based on the initial frequency spectrum and a target frequency band, a frequency spectrum refinement is performed on the target frequency band to generate an enhanced frequency spectrum;

[0016] Based on the enhanced frequency spectrum, a frequency domain feature matrix is constructed;

[0017] A graphical conversion is performed on the frequency domain feature matrix to generate the frequency domain feature map of the target ultrasonic sensor.

[0018] In some embodiments, the frequency domain feature map is input into a fusion feature network to generate a fusion feature map used for representing the flatness of the power battery cooling plate, including:

[0019] A channel attention weighting operation is performed on the frequency domain feature map to generate an enhanced feature map;

[0020] A feature splicing operation is performed on the enhanced feature map to generate a spliced feature map;

[0021] A spatial weight mapping operation is performed on the spliced feature map through a preset convolution kernel to generate a spatial weight matrix;

[0022] A weighted fusion operation is performed on the spatial weight matrix and the enhanced feature map to generate the fusion feature map used for representing the flatness of the power battery cooling plate.

[0023] In some embodiments, the flatness recognition model includes a convolution layer, a down-sampling layer, a global pooling layer, and a fully connected layer, the fusion feature map is input into the flatness recognition model, and a flatness detection result of the power battery cooling plate is output, including:

[0024] A feature channel conversion operation is performed on the fusion feature map through the convolution layer to generate a first feature map;

[0025] The first feature map is compressed in spatial dimension by a downsampling layer to generate a deep feature map.

[0026] A global pooling layer is used to compress the feature vectors of the deep feature map to generate compressed feature vectors.

[0027] The compressed feature vector is mapped by a fully connected layer to output the flatness detection result of the power battery cold plate.

[0028] In some implementations, it also includes:

[0029] Based on the comparison between the flatness test results and the preset flatness threshold, the production adjustment instructions for the power battery cold plate are determined.

[0030] Secondly, this application proposes a flatness detection device for a power battery cold plate, comprising:

[0031] A reflected signal receiving unit is used to acquire the reflected signal received by the target ultrasonic sensor after being reflected by the surface of the power battery cold plate when multiple ultrasonic sensors emit ultrasonic pulse signals toward the surface of the power battery cold plate. The multiple ultrasonic sensors include the target ultrasonic sensor.

[0032] The frequency domain spectrum generation unit generates a frequency domain feature spectrum of the target ultrasonic sensor based on the reflected signal;

[0033] The fusion feature generation unit is used to input the frequency domain feature map into the fusion feature network to generate a fusion feature map that characterizes the flatness of the power battery cold plate.

[0034] The detection result output unit is used to input the fused feature map into the flatness recognition model and output the flatness detection result of the power battery cold plate.

[0035] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the method for detecting the flatness of a power battery cold plate according to any one of the first aspects.

[0036] Fourthly, this application proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for detecting the flatness of the power battery cold plate according to any one of the first aspects.

[0037] Fifthly, this application proposes a computer program product, including a computer program, which, when executed by a processor, implements the method for detecting the flatness of a power battery cold plate according to any one of the first aspects.

[0038] In summary, the method for detecting the flatness of cold plates in power batteries provided in this application achieves efficient detection of cold plate flatness by emitting ultrasonic pulse signals to the surface of the cold plate and acquiring the reflected signals through multiple ultrasonic sensors. After frequency domain feature map generation, feature network processing, and flatness recognition model analysis, the method achieves efficient detection of cold plate flatness. This application utilizes the non-contact measurement characteristics of ultrasonic sensors, avoiding the time-consuming clamping and positioning problems of traditional coordinate measuring machines (CMMs), thus shortening the detection time. Through multi-point, multi-angle acoustic signal acquisition and frequency domain feature analysis, combined with automated processing of the feature network and recognition model, the comprehensiveness and accuracy of the detection data are improved, enabling precise characterization of cold plate flatness. Simultaneously, it eliminates the need for expensive CMM equipment, reducing detection costs and improving production efficiency, providing an efficient and low-cost solution for mass production quality inspection of power battery cold plates. Attached Figure Description

[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0040] Figure 1 A schematic flowchart of a method for detecting the flatness of a power battery cold plate provided in an embodiment of this application;

[0041] Figure 2 This is a schematic diagram of the structure of a power battery liquid cooling plate provided in an embodiment of this application;

[0042] Figure 3 A schematic diagram of a flatness detection device for a power battery cold plate provided in an embodiment of this application;

[0043] Figure 4 This is a schematic diagram of an electronic device for detecting the flatness of a power battery cold plate, provided in an embodiment of this application.

[0044] The correspondence between the reference numerals and component names in the figure is as follows:

[0045] 21 is the top cover plate, 22 is the flow channel plate, and 23 is the metal connector. Detailed Implementation

[0046] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0047] Please see Figure 1 This is a schematic flowchart of a method for detecting the flatness of a power battery cold plate according to an embodiment of this application, which specifically includes:

[0048] S110. When multiple ultrasonic sensors emit ultrasonic pulse signals toward the surface of the power battery cold plate, the reflected signal received by the target ultrasonic sensor and reflected by the surface of the power battery cold plate is acquired, wherein the multiple ultrasonic sensors include the target ultrasonic sensor.

[0049] For example, this step involves emitting ultrasonic pulse signals onto the surface of the power battery cold plate using multiple spatially distributed ultrasonic sensors. Utilizing the reflection characteristics of sound waves on a medium surface, the target sensor receives the reflected signals modulated by the surface topography of the cold plate. Because variations in the flatness of the cold plate cause differences in the sound wave reflection path, intensity, and phase, the reflected signals carry geometric feature information of local areas of the cold plate. The collaborative operation of multiple sensors can simultaneously cover different areas of the cold plate (such as the central area and edge areas), forming a spatially complementary data acquisition network.

[0050] Acquiring reflected signals is essentially the process of transforming the physical morphology of a cold plate into quantifiable acoustic data. Due to the differences in the propagation characteristics of ultrasound waves in planar and uneven regions (e.g., specular reflection is dominant in planar regions, while scattering occurs in uneven regions), the time-domain waveform and spectral characteristics of the reflected signal are directly related to the local flatness state. By simultaneously acquiring data from multiple sensors, an acoustic mapping model of the cold plate surface morphology can be constructed, providing raw input for subsequent frequency domain feature extraction and laying the data foundation for a comprehensive evaluation of flatness.

[0051] S120. Based on the reflected signal, generate the frequency domain feature map of the target ultrasonic sensor;

[0052] For example, generating a frequency domain feature map based on the reflected signal involves mapping the time-domain acoustic wave characteristics into a quantifiable frequency domain spatial expression. The reflected signal, as a physical representation of the interaction between ultrasound and the surface morphology of the cold plate, exhibits waveform distortions (such as amplitude attenuation and phase shift) directly related to the geometric features of the local region. By converting the time-domain signal into an initial spectrum using Fourier transform, the distribution pattern of acoustic wave energy along the frequency dimension is revealed. Since flatness anomalies can lead to energy concentration or scattering in specific frequency bands (e.g., narrow-band spectral peaks in planar regions, and broadband spectral diffusion in uneven regions), this step provides an analytical mathematical basis for flatness assessment.

[0053] The essence of frequency domain feature maps is a high-order visualization of spectral information. Specifically, firstly, a Chirp-Z transform is used to refine the spectrum, increasing the frequency resolution for the target frequency band. This allows for a more refined characterization of the blurred frequency components in the initial spectrum, amplifying the differences in frequency components under different planarity states. Then, based on the refined and enhanced spectrum, a frequency domain feature matrix coupling energy and phase is constructed. Through logarithmic amplitude compression (suppressing strong signals and enhancing the visibility of weak signals) and phase difference calculation between adjacent frequency points, the energy distribution characteristics and phase correlation properties of the spectrum are transformed into matrix elements. The values ​​of these matrix elements are determined by the logarithmic amplitude product, the cosine of the phase difference, and the frequency distance attenuation factor. Finally, through quantile normalization and gradient amplitude calculation, the matrix is ​​converted into a two-dimensional heatmap, visually presenting the coupling relationship between frequency, energy, and phase using color gradients and contour lines. This forms a machine-recognizable visualization carrier of planarity features, achieving a structured mapping from spectral data to planarity features.

[0054] S130. Input the frequency domain feature map into the fusion feature network to generate a fusion feature map for characterizing the flatness of the power battery cold plate;

[0055] For example, the frequency domain feature maps generated by each sensor are input into a fusion feature network to address the limitations of single-point detection. Since the flatness anomalies of the power battery cold plate typically manifest as regional deformations (such as edge warping or central depression), the frequency domain feature maps from a single sensor can only reflect the local state. By dynamically aggregating feature representations from multiple detection points through a fusion feature network, and utilizing channel attention mechanisms (such as SEBlock) to assign differentiated weights to feature maps at different locations, the contribution of key areas is enhanced, and noise interference is suppressed, thereby constructing a joint feature representation covering the entire cold plate area.

[0056] The resulting fusion map is not a simple superposition, but rather achieves feature interaction through spatial weight mapping (such as generating a spatial weight matrix through 1×1 convolution), forming a fused feature map characterizing the overall flatness. This map is essentially an integrated projection of the acoustic features of the cold plate surface morphology into the frequency domain, with its matrix elements relating to the energy and phase coupling relationships at different locations. The final output fused feature map serves as the input to the flatness recognition model, providing a high-dimensional feature carrier for end-to-end flatness determination.

[0057] S140. Input the fused feature map into the flatness recognition model and output the flatness detection result of the power battery cold plate.

[0058] For example, the flatness recognition model, serving as an end-to-end classification decision module, takes a fused feature map as input. This map is essentially a two-dimensional visualization matrix formed by spatial energy phase coupling of multi-sensor frequency domain features. The model automatically extracts flatness-related features from the map using a pre-trained deep convolutional network architecture. Examples include the concentrated frequency band energy distribution in flat areas of the cold plate surface, and the frequency band energy dispersion and gradient abrupt changes caused by uneven areas. Since the fused feature map integrates the acoustic features of the entire cold plate, the model can directly achieve automated discrimination from multi-dimensional visualization features to flatness states through multi-layer convolution and pooling operations, without relying on manually designed feature extraction rules.

[0059] The flatness detection result output by the model is essentially a quantitative assessment generated based on the mapping relationship between acoustic features and the morphology of the cold plate. The contour gradient distribution in the fused feature map (e.g., steep edges correspond to local warping, and gentle regions correspond to an ideal plane) is decoded by the model into discrete flatness levels (e.g., "flat / uneven") or continuous values ​​(e.g., warping in millimeters). This result directly reflects the degree of deviation of the cold plate surface from the ideal geometric plane, transforming macroscopic deformation, which is difficult to measure directly, into a computable acoustic pattern recognition problem.

[0060] In summary, this embodiment of the application utilizes multiple ultrasonic sensors to emit ultrasonic pulse signals onto the surface of the power battery cold plate and acquire the reflected signals. Through frequency domain feature map generation, fusion feature network processing, and flatness recognition model analysis, efficient detection of the cold plate flatness is achieved. This embodiment leverages the non-contact measurement characteristics of ultrasonic sensors, eliminating the need for workpiece clamping, positioning, and calibration, thus avoiding the time-consuming problem of point-by-point data acquisition by traditional coordinate measuring machines and significantly shortening the detection time. The spatial layout of multiple ultrasonic sensors with continuously zigzag-shaped vertical projection points on the cold plate surface enables multi-point, multi-angle acoustic signal acquisition. Combined with analog-to-digital conversion, Fourier transform, and spectrum refinement operations during frequency domain feature map generation, it can comprehensively capture the acoustic characteristics of different regions on the cold plate surface, providing rich and accurate data support for flatness detection. The fusion feature network, through… Operations such as channel attention weighting, feature stitching, and spatial weight mapping effectively aggregate frequency domain features from multiple sensors, enhance the feature contribution of key areas, and suppress noise interference. Then, an automated feature extraction and analysis is performed using a flatness recognition model that includes convolutional layers and downsampling layers to accurately characterize the flatness state of the cold plate. In addition, this embodiment does not require expensive coordinate measuring equipment, reducing equipment purchase and maintenance costs. It can quickly output detection results and determine production adjustment instructions based on preset thresholds, improving production efficiency and providing an efficient and low-cost solution for mass production quality inspection of power battery cold plates.

[0061] In some instances, the spatial arrangement of multiple ultrasonic sensors satisfies the following condition: the vertical projection points of the multiple ultrasonic sensors on the surface of the power battery cold plate are distributed in a continuous zigzag pattern.

[0062] For example, multiple ultrasonic sensors are arranged in a continuous zigzag pattern (e.g., a V-shaped array) on the surface of the power battery cold plate, with at least three sensors. Specifically, with the geometric center of the cold plate surface as a reference, the spatial position of each sensor above the cold plate satisfies the following: its vertical projection points are continuously arranged along a preset zigzag path, forming a non-uniform distribution network covering the edge and center areas of the cold plate. The turning angle of the zigzag path is dynamically adjusted according to the size of the cold plate to ensure that the spacing between projection points matches the flatness sensitivity of key areas of the cold plate (such as the battery module installation area). During signal acquisition, each sensor synchronously transmits ultrasonic pulse signals with a frequency range of 20kHz to 100kHz, a pulse width of 10μs, and a sampling rate of 200kHz. Through a multipath reflection mechanism, the target sensor receives the reflected signal modulated by the surface topography of the cold plate. The planar area mainly exhibits specular reflection, while the concave and convex areas produce scattering. The time-domain waveform distortion (amplitude attenuation, phase shift) of the reflected signal is directly related to the local flatness state. This layout uses a spatially complementary data acquisition network to simultaneously capture the differences in acoustic wave propagation characteristics in different areas of the cold plate, providing global acoustic feature input for the flatness recognition model.

[0063] In summary, the aforementioned spatial layout design improves the accuracy and efficiency of flatness detection through a continuous polygonal projection point distribution and a multi-sensor collaborative acquisition mechanism. The projection points cover the edges and key central areas of the cold plate, avoiding blind spots inherent in traditional single-point detection. This is particularly beneficial for highly sensitive areas such as the battery module installation area, enabling directional enhanced monitoring. The polygonal distribution creates differentiated acoustic wave incident angles (e.g., oblique incident angles in a V-shaped array), allowing for simultaneous acquisition of acoustic wave scattering characteristics from different directions within the same area through multi-path reflection, enhancing the ability to detect minute deformations. The non-uniform layout combined with a synchronous emission mechanism suppresses environmental noise and signal crosstalk, improving the signal-to-noise ratio of reflected signals. The projection point spacing and turning angles can be dynamically adjusted according to the cold plate size, adapting to the detection needs of different cold plate specifications and avoiding the need for redesigning the hardware layout. This design transforms the physical morphology of the cold plate into a spatially correlated acoustic feature network, providing high-resolution input for the fusion feature network and laying the data foundation for accurate flatness identification.

[0064] In some instances, frequency domain feature maps of the target ultrasonic sensor are generated based on the reflected signal, including:

[0065] The reflected signal is converted from analog to digital to generate a digital signal;

[0066] Perform a Fourier transform on the digital signal to generate the initial spectrum;

[0067] Based on the initial spectrum and the target frequency band, the target frequency band is refined to generate an enhanced spectrum;

[0068] Construct a frequency domain feature matrix based on the enhanced spectrum;

[0069] The frequency domain feature matrix is ​​graphically transformed to generate the frequency domain feature map of the target ultrasonic sensor.

[0070] For example, when performing analog-to-digital conversion on the reflected signal, the analog reflected signal is first sampled by a data acquisition card at a sampling rate of 200kHz. This sampling rate satisfies the Nyquist sampling theorem, ensuring accurate signal reconstruction. The time interval between each sampling point is set to 10ms to fully capture the waveform characteristics of the ultrasonic pulse. After sampling, the reflected signal is converted into a digital signal by the data acquisition card. This digital signal is stored in the computer as a discrete numerical sequence, providing standardized digital input for subsequent frequency domain analysis. Essentially, this process converts the analog acoustic wave signal modulated by the surface morphology of the cold plate into a calculable digital quantity, eliminating noise interference during analog signal transmission and laying the data foundation for frequency domain feature extraction.

[0071] When performing dynamic range normalization on digital signals, the signal sequence is first extracted and its absolute maximum value max|X| is calculated (i.e., the maximum absolute value of the sequence samples); based on the preset simulation parameter β (β is an adjustment coefficient greater than 0, used to control the proportion of the coordination quantity), the coordination adjustment quantity ε is derived (ε = β·max|X|, to avoid denominator abnormality when max|X| approaches zero, and to suppress noise amplification with amplitude close to zero); then, for each sample x of the digital signal... i Through formula x i The mapping is completed using / (max|X|+ε), achieving signal normalization. The above method eliminates DC bias while preserving signal trends, constructing a normalization scale using the absolute maximum value combined with a coordination quantity. This unifies the dynamic range of amplitudes at different detection points, providing stable and comparable input for subsequent frequency domain feature extraction, and adapting to the training and inference needs of the recognition model.

[0072] A windowed Fourier transform is performed on the normalized signal using a preset window function (such as the Hanning window or Blackman window). The purpose is to reduce spectral leakage and improve the accuracy of frequency domain analysis by leveraging the spectral shaping properties of the window function. The preset parameters of the window function (such as window length and overlap rate) are pre-set according to the frequency range and resolution requirements of the cold plate surface detection. By segmenting the time-domain signal into frames and multiplying them with the window function, the signal is made to satisfy the periodicity assumption in the time domain. Then, the time-domain signal is converted into a frequency-domain representation through Fourier transform, generating an initial spectrum containing the energy distribution of each frequency component. This process transforms the physical characteristics related to the flatness and uniformity of the cold plate surface into frequency-domain energy distribution characteristics, providing raw data support for subsequent spectrum refinement.

[0073] Spectrum refinement, based on the initial spectrum and a preset target frequency band, is a crucial operation for accurately capturing the characteristic frequencies of the cold plate surface by improving the frequency resolution of the target frequency band. Specifically, this step employs algorithms such as Chirp Z-transform or frequency domain interpolation to locally amplify the spectrum of the target frequency band, increasing the number of sampling points per unit frequency interval. This allows for a more refined characterization of the blurred characteristic frequency components in the initial spectrum. For example, if the initial spectrum has a frequency resolution of 10 Hz, refining the target frequency band (1000-2000 Hz) tenfold can improve the resolution to 1 Hz, thereby more accurately identifying characteristic frequency shifts or energy peak changes caused by abnormal surface flatness of the cold plate.

[0074] The calculation of logarithmic amplitude and phase difference values ​​for the enhanced spectrum is to transform the frequency domain energy distribution into a numerical representation suitable for constructing the feature matrix. The calculation of the logarithmic amplitude value compresses the dynamic range of the spectrum, making the difference between strong and weak signals more easily comprehensible numerically, while preserving the relative energy relationships of the spectrum. The calculation of the phase difference value reflects the phase change trend between adjacent frequency points and is closely related to the acoustic characteristics of the cold plate surface. When the cold plate surface has complete flatness, the phase change is regular, while incomplete flatness leads to abrupt phase changes. The calculation of these two parameters provides multidimensional features that combine energy distribution and phase characteristics for the subsequent construction of the frequency domain feature matrix.

[0075] A frequency domain feature matrix is ​​constructed based on logarithmic amplitude values, phase difference values, and a preset frequency distance attenuation factor. The attenuation factor weights the feature importance of different frequency points. The frequency distance attenuation factor is preset to be a function that decreases with increasing distance from the center frequency (such as an exponential attenuation function), aiming to suppress interference from distant frequencies and highlight characteristic frequency components related to the flatness of the cold plate. In the specific construction process, frequency is plotted on the horizontal axis, and logarithmic amplitude value and phase difference value are plotted on the vertical axis. The weighted logarithmic amplitude value and phase difference value of each frequency point are used as matrix elements to form a two-dimensional frequency domain feature matrix. This matrix not only preserves the energy distribution and phase information of the spectrum but also enhances the discriminative power of the features through the attenuation factor, providing structured data for subsequent spectrum generation.

[0076] Quantile normalization and gradient magnitude calculation of the frequency domain feature matrix are performed to eliminate the influence of outliers in the matrix and extract local variation features of the spectrum. Quantile normalization maps matrix elements to a preset quantile range (10% quantile and 90% quantile), giving the feature matrix of different detection points a uniform numerical distribution and enhancing feature contrast. Gradient magnitude calculation is obtained by solving the partial derivatives of the matrix in the row and column directions and calculating their Euclidean distance, which is used to extract the intensity information of matrix spatial variation.

[0077] Frequency domain feature maps are generated based on quantile normalization results and gradient magnitude calculation results, converting the numerical feature matrix into a visualized two-dimensional map to meet the input requirements of deep learning networks. Specifically, color mapping (such as heatmaps) maps the normalized logarithmic magnitude and gradient magnitude to the brightness and hue of colors, forming a visualized map that integrates energy distribution and variation features. Each pixel in the map corresponds to a frequency point, and the color features intuitively reflect the energy intensity and gradient variation at that frequency point. This allows the frequency domain features of the cold plate surface flatness (such as energy peak shifts and gradient abrupt change regions) to be presented visually, providing clear visual input features for subsequent map feature extraction networks.

[0078] In summary, this application's embodiments achieve precise conversion from reflected signals to visualized frequency domain features through multi-step processing including analog-to-digital conversion, Fourier transform, spectrum refinement, matrix construction, and graphical transformation. This method enhances the target frequency band resolution through Chirp-Z transform, capturing subtle frequency shifts that are difficult to identify using traditional methods. The energy-phase coupling matrix and frequency attenuation factor enhance the discriminative power of the features and effectively suppress noise interference. Quantile normalization and gradient superposition in the graphical transformation allow the spectrum to intuitively reflect flatness anomalies. Compared to the point-by-point data acquisition of traditional coordinate measuring machines, this application's embodiments shorten the single-sample processing time for cold plate flatness detection and improve detection accuracy, providing an efficient and high-resolution automated feature extraction solution for mass production quality inspection of power battery cold plates.

[0079] In some instances, the frequency domain feature map is input into a fusion feature network to generate a fusion feature map characterizing the flatness of the power battery cold plate, including:

[0080] Channel attention weighting is applied to the frequency domain feature map to generate an enhanced feature map.

[0081] Perform feature splicing operations on the enhanced feature map to generate a spliced ​​feature map;

[0082] By using a preset convolution kernel, a spatial weight mapping operation is performed on the spliced ​​feature map to generate a spatial weight matrix;

[0083] A weighted fusion operation is performed on the spatial weight matrix and the enhanced feature map to generate a fused feature map that characterizes the flatness of the power battery cold plate.

[0084] For example, when applying channel attention weighting to the frequency domain feature map, SEBlock (channel attention module) is used to filter the importance of feature channels. First, global average pooling is used to compress the spatial dimension of the frequency domain feature map into a global context description of the channel dimension, obtaining a global feature vector for each channel. Then, a fully connected layer is used to perform dimensionality reduction and expansion operations (e.g., first reducing the dimensionality to 1 / 4 of the original number of channels, then expanding it to the original number of channels) to generate channel weight vectors. Finally, the channel weight vectors are normalized to weight values ​​between 0 and 1 using the Sigmoid activation function and multiplied with the original frequency domain feature map to generate an enhanced feature map. This operation can suppress noise channels unrelated to flatness (such as environmental interference frequency bands) and strengthen key channels related to the surface morphology of the cold plate (such as frequency bands where reflection energy is concentrated), thereby improving the discriminative power of the feature map.

[0085] When performing feature stitching on the enhanced feature maps, multiple (e.g., three) enhanced feature maps generated by different ultrasonic sensors are stitched together along the channel dimension. Assuming each enhanced feature map has a size of H×W×C (H is the height, W is the width, and C is the number of channels), the stitched feature map generates an H×W×3C stitched feature map. This operation integrates frequency domain features acquired from multiple sensors from different angles, enabling the stitched feature map to simultaneously contain acoustic feature information from different locations on the cold plate surface, such as the central and edge regions, forming a joint feature representation covering the entire cold plate area. This provides multi-source feature input for subsequent spatial weight mapping, addressing the locality limitations of single-sensor detection.

[0086] When spatial weight mapping is performed on the spliced ​​feature map using a pre-defined convolutional kernel, the spliced ​​feature map has a specific spatial height, width, and total number of channels formed by splicing three-channel features. A 1×1 convolutional kernel with a stride of 1 is used to perform convolution operations on the spliced ​​feature map. Since the 1×1 convolutional kernel covers the features of all channels at each spatial position during its sliding process, a learnable weight matrix can be used to fuse the features of different channels at the same spatial position. The resulting spatial weight matrix has the same spatial height and width as the spliced ​​feature map. The value of each element in the matrix represents the correlation strength between the feature at the corresponding spatial position and the flatness of the cold plate. For example, due to distortion in the frequency domain features, the weight value of a warped area on the edge of the cold plate will be higher than that of a flat area. The above steps allow the convolutional network to autonomously learn the contribution of different spatial regions on the surface of the cold plate to the flatness detection results. By assigning higher weights to key regions, spatial selective enhancement of relevant features is achieved, providing precise weight guidance for subsequent feature aggregation.

[0087] When performing weighted fusion of the spatial weight matrix and the enhanced feature maps, the spatial weight matrix is ​​first multiplied element-wise with each enhanced feature map, allowing the weight matrix to adjust the spatial location features of the enhanced feature maps. Then, the multiple weighted enhanced feature maps are added along the channel dimension to generate the final fused feature map. In this process, the spatial weight matrix acts as an attention mask, highlighting the features of areas with abnormal flatness on the cold plate surface (such as spectral abrupt changes in local depressions) and suppressing interference information in non-critical areas. This allows the fused feature map to retain the original frequency domain features of each sensor while also aggregating features through spatial weights, ultimately forming a comprehensive feature map characterizing the overall flatness of the cold plate.

[0088] In summary, this application embodiment achieves efficient aggregation and optimization of multi-sensor frequency domain features through multi-step processing including channel attention weighting, feature stitching, spatial weight mapping, and weighted fusion. The channel attention mechanism automatically filters key frequency bands related to flatness, avoiding interference from invalid features. Multi-sensor feature stitching covers the acoustic features of the entire cold plate surface, solving the blind spot problem of single-point detection. The spatial weight mapping operation adaptively highlights features in areas of abnormal flatness, improving the accuracy of the fused feature map in representing local deformations (such as edge warping and central depressions). Compared to the point-by-point data integration method of traditional coordinate measuring machines, this application embodiment reduces the multi-sensor data fusion processing time from minutes to milliseconds and can capture minute changes in flatness, providing an efficient and robust feature fusion solution for online inspection of power battery cold plates in mass production.

[0089] In some instances, the flatness recognition model includes convolutional layers, downsampling layers, global pooling layers, and fully connected layers. The fused feature map is input into the flatness recognition model, and the output is the flatness detection result of the power battery cold plate, including:

[0090] The first feature map is generated by performing feature channel transformation on the fused feature map through a convolutional layer.

[0091] The first feature map is compressed in spatial dimension by a downsampling layer to generate a deep feature map.

[0092] The deep feature map is compressed using a global pooling layer to generate compressed feature vectors.

[0093] The compressed feature vector is mapped by a fully connected layer to output the flatness detection result of the power battery cold plate.

[0094] For example, the fused feature map is input into the convolutional layer of the flatness recognition model, and a feature channel transformation operation is performed using a convolutional kernel of a preset size (using a 3×3 convolutional kernel). This operation extracts spatial features and expands the channel dimension of the fused feature map. The convolutional kernel slides across the two-dimensional plane of the fused feature map to calculate the weighted sum of the local receptive fields. Through a preset number of output channels (initially converted to 64 channels), the multidimensional features of the input data are remapped to a higher-dimensional feature space. The output result is the first feature map, whose spatial size is consistent with the input fused feature map (e.g., H×W), but the channel dimension is expanded to the preset number of channels (e.g., 64 channels). This step enhances the feature representation capability through convolution operations, capturing local texture patterns related to flatness in the fused feature map (e.g., frequency domain energy gradient abrupt change regions), laying the foundation for deep feature extraction.

[0095] The first feature map is input into a downsampling layer (2×2 max pooling layer) to perform spatial dimension compression. This operation divides the input feature map into non-overlapping sub-regions with a preset stride (stride of 2), extracts the maximum value in each sub-region as the output, halving the spatial size of the feature map (e.g., from H×W to H / 2×W / 2), while preserving salient features and suppressing redundant details. After multi-stage cascaded operations (three stages), each stage further extracts features through convolutional layers and gradually compresses the spatial dimension in conjunction with the downsampling layer: Stage 1: Output size is compressed to 1 / 2 of the input size (e.g., H / 2×W / 2), maintaining 64 channels; Stage 2: Output size is compressed to 1 / 4 of the input size (e.g., H / 4×W / 4), expanding the number of channels to 128 channels; Stage 3: Output size is compressed to 1 / 8 of the input size (e.g., H / 8×W / 8), expanding the number of channels to 256 channels. The stepwise compression of spatial dimensions reduces computational complexity while preserving key features, and enhances the translation invariance of features, making the model robust to small displacements caused by deformation of the cold plate surface.

[0096] The final deep feature map (H / 8×W / 8×256) is input into a global average pooling layer to perform feature vector compression. This operation calculates the average value of all spatial locations for each channel of the feature map, compressing the two-dimensional feature matrix into a one-dimensional vector (e.g., compressing H / 8×W / 8×256 into a 1×1×256 dimensional vector). This step completely eliminates the spatial dimension, preserves the global statistical characteristics of each channel, and generates a high-information-density compressed feature vector. This reduces the parameter scale of subsequent fully connected layers to avoid overfitting and also aggregates the acoustic feature distribution patterns of the entire cold plate (e.g., the frequency domain energy difference between the central and edge regions).

[0097] The compressed feature vector is input into a fully connected layer, where a fully connected mapping operation is performed. This operation uses a predetermined number of fully connected neurons to perform nonlinear transformations and dimensional mapping on the compressed feature vector. Specifically: first, the 256-dimensional vector is expanded to 512 dimensions through the fully connected layer to enhance feature interaction capabilities; then, it is mapped to the output dimension corresponding to the flatness classification result through the final fully connected layer (e.g., a 2-dimensional node for binary classification, or an N-dimensional node for multi-level scoring). The fully connected layer learns the complex relationship between the compressed feature vector and the flatness state through the weight matrix, ultimately outputting a discrete flatness level (e.g., "flat / unflat") or a continuous value (e.g., warpage in millimeters), directly representing the degree of deviation of the cold plate surface from the ideal geometric plane.

[0098] In summary, this application's embodiments achieve end-to-end intelligent analysis from fused feature maps to flatness detection results through a cascaded design of convolutional layers, downsampling layers, global pooling layers, and fully connected layers. The model automatically extracts local features related to flatness using convolutional layers, avoiding the subjectivity and limitations of manually designed features in traditional methods. The combination of downsampling and global pooling preserves key features while reducing computational complexity, enabling the model to process multi-sensor fusion data in real-time in industrial settings. The nonlinear mapping capability of the fully connected layers captures the complex correlation between flatness anomalies and frequency domain features. Furthermore, the model does not rely on expensive 3D scanning equipment; it achieves automated flatness evaluation through software algorithms, improving the quality inspection efficiency of mass production of power battery cold plates and reducing equipment investment and maintenance costs for enterprises.

[0099] In some instances, it also includes:

[0100] Based on the comparison between the flatness test results and the preset flatness threshold, the production adjustment instructions for the power battery cold plate are determined.

[0101] It should be noted that the power battery cooling plate, as a core component of the thermal management system, can be divided into several types based on differences in cooling methods and structures. Liquid-cooled plates achieve heat exchange through internal flow channels and coolant convection; direct-cooled plates utilize the latent heat of refrigerant phase change to achieve heat conduction with the power battery; and harmonica-style cooling plates improve heat dissipation uniformity through a parallel pipe structure. This patent uses a liquid-cooled plate as an example; please refer to [link / reference]. Figure 2 This is a schematic diagram of the structure of a power battery liquid cooling plate provided in this application embodiment. It consists of, but is not limited to, an upper cover plate 21, a flow channel plate 22, and a metal connector 23. The upper cover plate 21 serves as the direct contact surface between the liquid cooling plate and the power battery module, and its flatness directly determines the contact area and heat transfer efficiency between the two. If the flatness of the upper cover plate 21 is insufficient, on the one hand, it will lead to a reduction in the contact area with the power battery module and a decrease in heat transfer efficiency; on the other hand, it will cause poor flow of coolant in certain areas, resulting in uneven battery temperature field, causing local overheating or overcooling, and even affecting the safety of the power battery. The flow channel plate 22 has a complex coolant flow channel inside, which is welded to the upper cover plate 21 to form a closed cavity. Its structural design affects the coolant flow rate distribution and heat dissipation uniformity, but the flatness detection of the power battery cooling plate mainly targets the surface of the upper cover plate, detecting the flatness of the upper cover plate 21. The metal connector 23 is used to connect external coolant pipes to realize the circulation of the thermal management system.

[0102] For example, when determining production adjustment instructions based on the comparison between the flatness test results and the preset flatness threshold, the flatness threshold must first be preset according to the design specifications and application scenarios of the power battery cold plate. For example, in the natural state, the overall flatness threshold of the upper cover plate of the cold plate is set to ≤10mm, and the area where the power battery module is placed is set to ≤5mm. In the pressed state, the overall flatness threshold of the upper cover plate is set to ≤1.4mm, and the area where the power battery module is placed is set to ≤0.8mm. This threshold can be flexibly adjusted according to the type of cold plate (such as liquid cooling plate, direct cooling plate) and the voltage level of the assembled battery pack (such as 400V, 800V). This patent does not make specific limitations. After the flatness recognition model outputs the detection result, it is compared with the corresponding threshold: if the detection result is less than or equal to the threshold, the cold plate is deemed to be flat and qualified, and a production instruction of "allowed to flow into the next process" is generated; if the detection result exceeds the threshold, the area and degree of flatness abnormality are further analyzed. For example, the specific location of edge warping or center concavity is located by fusing feature maps, and a targeted adjustment instruction is generated based on the cold plate's usage scenario. That is, for cold plates with slight deviations (such as flatness deviation within 0.5mm), a "concession acceptance" instruction is generated and marked "test part", indicating that it does not meet the conditions for loading. "Acceptance" indicates that products exceeding the standard are accepted to a limited extent, rather than being fully recognized as meeting quality standards. The label "Test Part" clarifies the purpose of the cold plate, which can only be used for R&D testing, process verification, and other experimental scenarios, and must not be assembled into the officially manufactured power battery pack. For cold plates with serious deviations (such as flatness deviation > 1mm) or deviations in critical areas (module installation area), a "Scrap Disposal" instruction is generated, and the deviation data is fed back to the production process system. The flatness of the cold plate can be ensured to meet the assembly and heat conduction requirements of the power battery thermal management system by checking the quality of incoming materials or adjusting the speed of the production line conveyor belt.

[0103] It should be noted that the natural state refers to the power battery cold plate being free from any external force, while the compressed state refers to the power battery cold plate bearing load pressure. Preferably, the load pressure is 2 kPa, but it can also be flexibly adjusted according to the actual situation. This patent does not impose specific limitations.

[0104] Please see Figure 3 The diagram below illustrates a flatness detection device for a power battery cold plate, as provided in an embodiment of this application. The device includes:

[0105] The reflected signal receiving unit 201 is used to acquire the reflected signal received by the target ultrasonic sensor after being reflected by the surface of the power battery cold plate when multiple ultrasonic sensors emit ultrasonic pulse signals toward the surface of the power battery cold plate. The multiple ultrasonic sensors include the target ultrasonic sensor.

[0106] The frequency domain spectrum generation unit 202 generates a frequency domain feature spectrum of the target ultrasonic sensor based on the reflected signal;

[0107] The fusion feature generation unit 203 is used to input the frequency domain feature map into the fusion feature network to generate a fusion feature map for characterizing the flatness of the power battery cold plate.

[0108] The detection result output unit 204 is used to input the fused feature map into the flatness recognition model and output the flatness detection result of the power battery cold plate.

[0109] Please see Figure 4 This application also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of a method for detecting the flatness of a power battery cold plate.

[0110] Since the electronic device described in this embodiment is the device used to implement the flatness detection device for a power battery cold plate in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any device used by those skilled in the art to implement the method in this application embodiment is within the scope of protection of this application.

[0111] In practice, when the computer program 311 is executed by the processor, it can implement any of the embodiments corresponding to the first aspect.

[0112] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0113] Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media containing computer-readable program code.

[0114] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to perform... Figure 1 The flowchart of a method for detecting the flatness of a power battery cold plate in the corresponding embodiment.

[0118] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, computer instructions may be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium may be any usable medium that a computer can store or a data storage device such as a server or data center that integrates one or more usable media. The usable medium may be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0121] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0122] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in the form of hardware and / or software functional units.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, magnetic disks, or optical disks.

[0124] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0125] Although preferred embodiments have been described in this specification, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications that fall outside the scope of this specification.

[0126] Obviously, those skilled in the art can make various modifications to this specification without departing from its spirit and scope. Therefore, this specification also intends to include any modifications that fall within the scope of the claims and their equivalents.

Claims

1. A method for detecting the flatness of a power battery cold plate, characterized in that, include: When multiple ultrasonic sensors emit ultrasonic pulse signals toward the surface of the power battery cold plate, the reflected signal received by the target ultrasonic sensor after being reflected by the surface of the power battery cold plate is acquired, wherein the multiple ultrasonic sensors include the target ultrasonic sensor. Based on the reflected signal, a frequency domain feature map of the target ultrasonic sensor is generated; The frequency domain feature map is input into the fusion feature network to generate a fusion feature map for characterizing the flatness of the power battery cold plate; The fused feature map is input into the flatness recognition model, and the flatness detection result of the power battery cold plate is output.

2. The method according to claim 1, characterized in that, The spatial arrangement of the multiple ultrasonic sensors satisfies the following condition: the vertical projection points of the multiple ultrasonic sensors on the surface of the power battery cold plate are distributed in a continuous zigzag pattern.

3. The method according to claim 1, characterized in that, The step of generating the frequency domain feature map of the target ultrasonic sensor based on the reflected signal includes: The reflected signal is subjected to analog-to-digital conversion to generate a digital signal; Perform a Fourier transform on the digital signal to generate an initial spectrum; Based on the initial spectrum and the target frequency band, the target frequency band is refined to generate an enhanced spectrum; Based on the enhanced spectrum, a frequency domain feature matrix is ​​constructed; The frequency domain feature matrix is ​​graphically transformed to generate the frequency domain feature spectrum of the target ultrasonic sensor.

4. The method according to claim 1, characterized in that, The step of inputting the frequency domain feature map into the fusion feature network to generate a fusion feature map for characterizing the flatness of the power battery cold plate includes: Channel attention weighting is applied to the frequency domain feature map to generate an enhanced feature map; Perform feature splicing operation on the enhanced feature map to generate a spliced ​​feature map; By using a preset convolution kernel, a spatial weight mapping operation is performed on the spliced ​​feature map to generate a spatial weight matrix; A weighted fusion operation is performed on the spatial weight matrix and the enhanced feature map to generate the fused feature map used to characterize the flatness of the power battery cold plate.

5. The method according to claim 1, characterized in that, The flatness recognition model includes convolutional layers, downsampling layers, global pooling layers, and fully connected layers. The step of inputting the fused feature map into the flatness recognition model and outputting the flatness detection result of the power battery cold plate includes: The convolutional layer performs feature channel transformation on the fused feature map to generate a first feature map. The first feature map is compressed in spatial dimension by the downsampling layer to generate a deep feature map. The global pooling layer is used to perform feature vector compression on the deep feature map to generate compressed feature vectors. The compressed feature vector is mapped by the fully connected layer to output the flatness detection result of the power battery cold plate.

6. The method according to claim 1, characterized in that, Also includes: Based on the comparison between the flatness detection result and the preset flatness threshold, the production adjustment instruction for the power battery cold plate is determined.

7. A flatness detection device for a power battery cold plate, characterized in that, include: A reflected signal receiving unit is used to acquire the reflected signal received by the target ultrasonic sensor after being reflected by the surface of the power battery cold plate when multiple ultrasonic sensors emit ultrasonic pulse signals toward the surface of the power battery cold plate, wherein the multiple ultrasonic sensors include the target ultrasonic sensor. The frequency domain spectrum generation unit generates a frequency domain feature spectrum of the target ultrasonic sensor based on the reflected signal; A fusion feature generation unit is used to input the frequency domain feature map into the fusion feature network to generate a fusion feature map for characterizing the flatness of the power battery cold plate. The detection result output unit is used to input the fused feature map into the flatness recognition model and output the flatness detection result of the power battery cold plate.

8. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program stored in the memory to implement the steps of the method for detecting the flatness of a power battery cold plate as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for detecting the flatness of the power battery cold plate as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for detecting the flatness of the power battery cold plate as described in any one of claims 1 to 6.

Images