CN114581446B - Battery core abnormity detection method and system of laminated battery - Google Patents
Battery core abnormity detection method and system of laminated battery Download PDFInfo
- Publication number
- CN114581446B CN114581446B CN202210485397.2A CN202210485397A CN114581446B CN 114581446 B CN114581446 B CN 114581446B CN 202210485397 A CN202210485397 A CN 202210485397A CN 114581446 B CN114581446 B CN 114581446B
- Authority
- CN
- China
- Prior art keywords
- image
- end point
- plate end
- negative plate
- positive plate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N23/00—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00
- G01N23/02—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material
- G01N23/04—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and forming images of the material
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2223/00—Investigating materials by wave or particle radiation
- G01N2223/10—Different kinds of radiation or particles
- G01N2223/101—Different kinds of radiation or particles electromagnetic radiation
- G01N2223/1016—X-ray
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2223/00—Investigating materials by wave or particle radiation
- G01N2223/40—Imaging
- G01N2223/401—Imaging image processing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P70/00—Climate change mitigation technologies in the production process for final industrial or consumer products
- Y02P70/50—Manufacturing or production processes characterised by the final manufactured product
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for detecting the abnormity of a battery core of a laminated battery, wherein the method comprises the following steps: acquiring an image of the battery cell to be detected through X-Ray equipment; preprocessing an image of the battery cell to be detected, and extracting an image of an interested area; inputting the region-of-interest image into a pre-established neural network detection model, and obtaining the number of layers of the positive plate and the negative plate in the region-of-interest image; segmenting the image of the region of interest according to the number of layers to obtain a segmented image; inputting the segmentation images into a pre-established convolution neural network model to obtain positive plate end point thermodynamic diagrams and negative plate end point thermodynamic diagrams in each segmentation image; screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram; performing abnormity detection based on the screened positive plate end point and the screened negative plate end point; the method calculates the number of the pole piece layers through the neural network, and performs feature recognition through the convolutional neural network model, so that the anomaly detection accuracy is high and the efficiency is high.
Description
Technical Field
The invention relates to the technical field of battery core detection, in particular to a battery core abnormity detection method and system of a laminated battery.
Background
The method for carrying out nondestructive testing on the laminated lithium ion battery cell by using the X-ray imaging technology becomes an essential link in production, and is favorable for quality control of the battery. When the lithium electronic battery cell is detected to be abnormal by adopting X rays, the X rays are firstly sent out by an X-Ray emitter to penetrate through the battery cell, then the X rays are received by a receiving terminal and imaged, then the X-Ray image is processed by a software algorithm to obtain related data, and finally, the good product and the defective product are determined according to the process requirements. In the detection process, the automatic mechanical equipment is mainly responsible for logistics transportation, photographing positioning, NG blanking and the like of the battery cell or a battery consisting of the battery cell; and the software algorithm is used for judging good products and defective products through the X-Ray images of the battery cell, so that the whole detection is successfully completed. At present, the development of automatic mechanical equipment is mature, so the improvement of a software algorithm is particularly critical to the improvement of a detection effect.
The existing X-ray image processing mostly adopts the traditional image processing method, such as binarization and the like, which is difficult to deal with the complex situation, and the detection time is long, thus affecting the production efficiency. The battery usually comprises two or more electric cores, and has the shell parcel, is different from electric core in the formation of image, and the detection degree of difficulty to the battery is higher.
For example, patent document CN104091322A discloses a method for detecting a laminated lithium ion battery, which starts with an X-Ray grayscale image of the laminated battery, determines a feature region by using a statistical method, sets a threshold to detect corner points of the feature region, and then performs screening, compensation and fitting on positive and negative corner points with the highest probability, and performs abnormality detection on the basis of the above steps in sequence.
The method can be used for carrying out safety detection on the internal structure of the lithium battery through an X-Ray visual system, but the detection time is long, and the detection accuracy is low and detection omission occurs when the condition is complex.
Disclosure of Invention
The invention provides a method and a system for detecting the battery core abnormity of a laminated battery.
A method for detecting the abnormity of a battery core of a laminated battery comprises the following steps:
acquiring an image of the battery cell to be detected through X-Ray equipment;
preprocessing the to-be-detected battery cell image, and extracting an interested area image;
inputting the region-of-interest image into a pre-established neural network detection model to obtain the number of layers of the positive plate and the negative plate in the region-of-interest image;
segmenting the image of the region of interest according to the layer number to obtain a segmented image;
inputting the segmentation images into a pre-established convolutional neural network model to obtain positive plate end point thermodynamic diagrams and negative plate end point thermodynamic diagrams in each segmentation image;
screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
and carrying out abnormity detection based on the screened positive plate end point and the screened negative plate end point.
Further, the step of preprocessing the image of the to-be-detected battery cell and extracting the image of the region of interest includes:
carrying out normalization processing on the to-be-detected battery cell image to obtain an 8-bit gray level image;
extracting boundary points of the 8-bit gray level image according to a set gray level threshold value;
and determining boundary points and two side boundary points of one end part of the positive plate and the negative plate in the 8-bit gray level image, and intercepting an area containing the boundary points and the two side boundary points of one end part of the positive plate and the negative plate as the region-of-interest image.
Further, when the number of layers of the positive plate and the negative plate in the region-of-interest image is obtained, the method further includes:
and identifying a single-side sheet of the battery cell through the neural network detection model, wherein the single-side sheet is used for distinguishing two or more battery cells.
Further, the convolutional neural network model includes a first convolutional layer, a first residual module, a first maximum pooling layer, a second residual module, a second maximum pooling layer, a second convolutional layer, and a third convolutional layer.
Further, inputting the segmentation images into a pre-established convolutional neural network model to obtain a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram in each segmentation image, and the method comprises the following steps:
inputting the segmentation image into the first convolution layer to carry out 3-by-3 convolution with the step length of 2 to obtain a first feature map;
inputting the first feature map into the first residual error module and the first maximum pooling layer, and respectively calculating to obtain a second feature map;
inputting the second feature map into the second residual error module and a second maximum pooling layer, and respectively calculating to obtain a third feature map;
after the third feature map is subjected to upsampling, the third feature map is spliced with the second feature map and is input into the second convolution layer to be subjected to 1-by-1 convolution, and a convolution result is obtained;
and after the convolution result is subjected to up-sampling, the convolution result is spliced with the first characteristic diagram and then input to the third convolution layer for 1-1 convolution, and a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram are obtained.
Further, establishing a convolutional neural network model, comprising:
establishing an initial convolutional neural network model;
collecting a pole piece image sample, and processing the pole piece image sample to enable the pixel values of a positive pole piece endpoint and a negative pole piece endpoint in the pole piece sample image to be 1;
performing Gaussian filtering on the processed pole piece image sample to enable the pixel value far away from the positive pole piece end point and the negative pole piece end point in the pole piece image sample to be gradually reduced to 0;
and inputting the pole piece image sample subjected to Gaussian filtering into the initial convolutional neural network model for training, calculating the value of a loss function in the training, and obtaining the convolutional neural network model when the value of the loss function is smaller than a preset value.
Further, the loss function is a mean square error loss function.
Further, the pixel value of each pixel point in the positive plate endpoint thermodynamic diagram is the probability that the position is the positive plate endpoint, and the pixel value of each pixel point in the negative plate endpoint thermodynamic diagram is the probability that the position is the negative plate endpoint;
screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram, and the screening method comprises the following steps:
carrying out non-local maximum suppression on pixels in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
selecting pixel points with pixel values larger than the pixel threshold value according to the set pixel threshold value;
for the negative plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k-1, selecting the pixels with the largest pixel values from the rest pixels until the number of the selected pixels reaches k-1;
for the positive plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k, selecting the pixels with the largest pixel values in the rest pixels until the number of the selected pixels reaches k;
and k is the number of layers of the positive plates in the segmentation image.
Further, the abnormality detection based on the screened positive plate end point and the screened negative plate end point includes:
and calculating the layer number, the coating value, the fall and the bending angle of the positive plate and the negative plate according to the position coordinates of the end points of the positive plate and the negative plate, and judging whether the battery cell is abnormal.
A battery core abnormity detection system of a laminated battery comprises an X-Ray device, a processor and a storage device, wherein the storage device stores a plurality of instructions, and the processor is used for reading the instructions and executing the method.
The method and the system for detecting the battery core abnormity of the laminated battery at least have the following beneficial effects:
(1) before the characteristic recognition is carried out on the image, besides the binaryzation extraction boundary point, the layer number of the positive plate and the layer number of the negative plate are calculated through the neural network model, and then the image is divided into a plurality of smaller divided images serving as input data of the convolutional neural network model according to the calculated layer number, so that the recognition accuracy is better when the complex detection condition is met, and the operation efficiency is improved;
(2) the pole piece characteristics are identified through a pre-trained convolutional neural network model to obtain a thermodynamic diagram so as to determine the end point position of the pole piece, and compared with the traditional methods such as setting a threshold value and the like, the method has better identification accuracy and greatly reduces the false detection rate;
(3) the single-surface sheet is identified while the positive plate and the negative plate are identified, the plurality of battery cells are distinguished through the single-surface sheet, and the abnormity detection of the battery is simplified into the abnormity detection of the battery cells, so that the method can be used for carrying out abnormity detection on the battery comprising the plurality of battery cells;
(4) the convolution neural network model of the output pixel value representing endpoint probability is established, and the endpoints of the positive plate and the negative plate are determined in a probability judgment and pixel point screening mode.
Drawings
Fig. 1 is a schematic structural diagram of a battery cell for detecting the battery cell abnormality detection method of the laminated battery provided by the invention.
Fig. 2 is a flowchart of an embodiment of a method for detecting an abnormal cell state of a laminated battery according to the present invention.
Fig. 3 is a schematic diagram illustrating a method for detecting a cell abnormality of a laminated battery according to an embodiment of the present invention, in which an image of a region of interest is extracted.
Fig. 4 is a schematic diagram illustrating a single-surface sheet identification according to an embodiment of the method for detecting abnormal cell state of a laminated battery provided by the present invention.
Fig. 5 is a result schematic diagram of an embodiment of a convolutional neural network model in the method for detecting the cell abnormality of the laminated battery provided by the present invention.
Fig. 6 is a flowchart of an embodiment of a method for obtaining a thermodynamic diagram through a convolutional neural network model in an anomaly detection method provided by the present invention.
Fig. 7 is a schematic diagram of a positive plate end point thermodynamic diagram according to an embodiment of the method for detecting battery core abnormality of a laminated battery provided by the present invention.
Fig. 8 is a schematic diagram of a negative plate end point thermodynamic diagram according to an embodiment of the method for detecting battery core abnormality of a laminated battery provided by the present invention.
Fig. 9 is a schematic structural diagram of an embodiment of a cell abnormality detection apparatus of a laminated battery provided in the present invention.
Fig. 10 is a schematic structural diagram of an embodiment of a cell abnormality detection system of a laminated battery provided in the present invention.
Description of the drawings: 1-processor, 1001-positive plate, 1002-negative plate, 1003-single plate, 1004-positive plate, 1005-negative plate, 101-acquisition module, 102-preprocessing module, 103-layer number calculation module, 104-segmentation module, 105-feature identification module, 106-screening module, 107-anomaly detection module, 2-storage device, 3-X-Ray equipment, 301-region-of-interest image, 401-single plate detection frame, 501-first rolling layer, 502-first residual module, 503-first maximum pooling layer, 504-second residual module, 505-second maximum pooling layer, 506-second rolling layer, 507-third rolling layer.
Detailed Description
In order to better understand the technical scheme, the technical scheme is described in detail in the following with reference to the attached drawings of the specification and specific embodiments.
To facilitate an understanding of the present application, a description will be given of a structure of a laminated battery cell to which the present application relates. Referring to fig. 1, a cell of a laminated battery includes a plurality of positive plates 1001 and negative plates 1002 arranged at intervals, the positive plates 1001 are located on the inner side, the negative plates 1002 are located on the outer side, a positive tab 1004 is led out from each positive plate 1001, and a negative tab 1005 is led out from each negative plate 1002. A separator is provided between the positive electrode sheet 1001 and the negative electrode sheet 1002, and lithium ions in the electrolyte in the separator move to generate electricity. A single-sided sheet 1003 is arranged between every two battery cells and used for distinguishing two or more battery cells.
The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.
Referring to fig. 2, in some embodiments, there is provided a cell abnormality detection method of a laminated battery, including:
s1, acquiring an image of the battery cell to be detected through X-Ray equipment;
s2, preprocessing the to-be-detected battery cell image, and extracting an interested area image;
s3, inputting the region-of-interest image into a pre-established neural network detection model, and obtaining the layer number of the positive plate and the negative plate in the region-of-interest image;
s4, segmenting the region-of-interest image according to the layer number to obtain a segmented image;
s5, inputting the segmentation images into a pre-established convolution neural network model to obtain positive plate end point thermodynamic diagrams and negative plate end point thermodynamic diagrams in each segmentation image;
s6, screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
and S7, carrying out abnormity detection based on the screened positive plate end point and the screened negative plate end point.
Specifically, the to-be-detected cell image acquired by the X-Ray device in step S1 is a 16-bit image.
In step S2, preprocessing the to-be-detected cell image, and extracting an image of an area of interest, including:
s21, carrying out normalization processing on the to-be-detected cell image to obtain an 8-bit gray level image;
s22, extracting boundary points of the 8-bit gray level image according to a set gray level threshold value;
s23, determining boundary points and two side boundary points of one end part of the positive plate and the negative plate in the 8-bit gray scale image, and intercepting an area containing the boundary points and the two side boundary points of one end part as the interested area image.
In step S22, binarization processing is performed on the 8-bit grayscale image according to a set grayscale threshold, so as to extract boundary points of the to-be-detected cell image.
In step S23, first, two side regions and one end region are identified in the electric core image to be measured, and then a series of boundary points included in the two side regions and the one end region are obtained, so as to intercept the region-of-interest image according to the boundary points and the preset offset. Referring to fig. 3, a region-of-interest image 301 expanded in size by the amount of offset is cut out from boundary points a and C on both sides and a boundary point B at one end portion thereof. Note that the offset amount is a pixel size that expands outward with the boundary point as a center. When an interested area is defined according to the series of boundary points, an expanded rectangle is drawn as an interested area image according to the three boundary points and with the offset as the expanded pixel size, so that the obtained interested area image is ensured to include all the end parts of the battery cell.
In step S3, the pre-trained neural network detection model may detect the end points of the positive plate and the negative plate, and further determine the number of layers of the positive plate and the negative plate.
As a preferred embodiment, when the number of layers of the positive electrode plate and the negative electrode plate in the region-of-interest image is obtained, the method further includes: and identifying a single-side sheet of the battery cell through a neural network detection model, wherein the single-side sheet is used for distinguishing two or more battery cells. Referring to fig. 1 and 4, a single-sided sheet 1003 between two battery cells is disposed in the single-sided sheet detection frame 401, and two battery cells on two sides of the single-sided sheet can be distinguished by one single-sided sheet.
When a battery including a plurality of battery cells is detected, the battery cells need to be first divided before the battery cells are detected, and then abnormality detection needs to be performed on each of the battery cells. The single-surface sheets are two cathode sheets on the outermost side of one battery cell, and two battery cells can be divided as long as the position of the single-surface sheet fit between the two battery cells is accurately positioned. Through the identification of a single surface sheet of the battery cell, the abnormity detection of the battery is simplified into the abnormity detection of the battery cell, so that the nondestructive detection of the whole lithium ion battery is realized by adopting an X-Ray image.
In step S4, the number of layers of positive plates and the number of layers of negative plates in each segmented image are first set, and then the image of the region of interest is divided into a plurality of smaller segmented images according to the number of layers of positive plates and negative plates in the image of interest, the width of each segmented image is determined by the number of layers of positive plates and the number of layers of negative plates in each segmented image, and the height of each segmented image is a preset multiple of the width. As a preferred embodiment, it is set that each of the divided images includes k positive electrode sheets, the height of the divided image is 2 times the width, and if the number of positive electrode sheets in the region-of-interest image is N, the number of the divided images obtained is N/k.
In step S5, referring to fig. 5, the convolutional neural network model includes a first convolutional layer 501, a first residual module 502, a first maximum pooling layer 503, a second residual module 504, a second maximum pooling layer 505, a second convolutional layer 506, and a third convolutional layer 507.
Referring to fig. 5 and 6, inputting the segmented images into a pre-established convolutional neural network model to obtain a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram in each segmented image, including:
s51, inputting the segmented image X into the first convolution layer 501 for 3 × 3 convolution with a step size of 2 to obtain a first feature map X1;
s52, inputting the first feature map X1 to the first residual module 502 and the first maximum pooling layer 503, and calculating to obtain a second feature map X2;
s53, inputting the second feature map X2 into the second residual module 504 and the second maximum pooling layer 505 to be calculated respectively, and then obtaining a third feature map X3;
s54, the third feature map X3 is upsampled, then is spliced with the second feature map, and is input to the second convolution layer 506 to perform 1 × 1 convolution, so as to obtain a convolution result X4;
and S55, after the convolution result X4 is subjected to upsampling and then spliced with the first characteristic map X1, inputting the result to the third convolution layer 507 for 1-by-1 convolution, and obtaining a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram.
Specifically, in step S51, the segmented image is input to the first convolution layer and is convolved by 3 × 3 with a step size of 2, so as to obtain a first feature map, and the width and height of the first feature map are reduced to 1/2, which is the size of the original segmented image. In step S52, the second feature map width and height are reduced to 1/4 of the original divided image size. In step S53, the third feature map width and height are reduced to 1/8 of the original divided image size. In step S54, the third feature map is up-sampled and then has the same size as the second feature map, and after being spliced and convolved with the second feature map, the width and height of the obtained convolution result map is 1/4 of the original segmentation image size. In step S55, the convolution result is the same as the first feature map in size after being upsampled, and after being spliced and convolved with the first feature map, a final convolution result map is obtained, which is the same in size as the original segmentation image. The final convolution result graph comprises two channels, wherein one channel is a positive plate end point thermodynamic diagram, and the other channel is a negative plate end point thermodynamic diagram. Fig. 7 is a schematic diagram of a positive plate terminal thermodynamic diagram, and fig. 8 is a schematic diagram of a negative plate terminal thermodynamic diagram.
Wherein, establishing the convolutional neural network model comprises:
establishing an initial convolutional neural network model;
collecting a pole piece image sample, and processing the pole piece image sample to enable the pixel values of a positive pole piece endpoint and a negative pole piece endpoint in the pole piece sample image to be 1;
performing Gaussian filtering on the processed pole piece image sample to enable the pixel values far away from the positive pole piece endpoint and the negative pole piece endpoint in the pole piece image sample to gradually decrease to 0;
and inputting the pole piece image sample subjected to Gaussian filtering into the initial convolutional neural network model for training, calculating the value of a loss function in the training, and obtaining the convolutional neural network model when the value of the loss function is smaller than a preset value.
Wherein the loss function is a mean square error loss function:
and M is the number of pole piece image samples, y is a true value, and x is a predicted value.
The pixel value of each pixel point in the positive plate endpoint thermodynamic diagram is the probability that the position is the positive plate endpoint, and the pixel value of each pixel point in the negative plate endpoint thermodynamic diagram is the probability that the position is the negative plate endpoint.
Specifically, the pixel values of the positive plate end point and the negative plate end point in the pole piece sample image are set to be 1, then the Gaussian filtering is performed to gradually reduce the pixel values far away from the positive plate end point and the negative plate end point in the pole piece image sample to be 0, so that the pixel values of the positive plate end point and the negative plate end point in the pole piece image sample are the maximum value 1, and the pixel values far away from the positive plate end point and the negative plate end point are gradually reduced to be 0, so that the pixel values can represent the probability value that the pixel is the end point, and then the probability that the pixel point is the pole piece end point can be detected through the convolutional neural network model.
In a preferred embodiment, the segmented image is input into a convolutional neural network model established in advance in the GPU for the characteristic identification operation, and the obtained positive plate end point thermodynamic diagrams and negative plate end point thermodynamic diagrams are output to the CPU for subsequent processing of the thermodynamic diagrams. The GPU has high operation speed, and can save a large amount of time by processing large-batch operation, so that the operation efficiency can be improved by carrying out feature recognition operation in the GPU.
And detecting a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram through a convolutional neural network model, wherein the pixel value of each pixel point in the positive plate end point thermodynamic diagram is the probability that the position is the positive plate end point, and the pixel value of each pixel point in the negative plate end point thermodynamic diagram is the probability that the position is the negative plate end point.
In step S6, the screening of the positive electrode tab end points and the negative electrode tab end points in the positive electrode tab end point thermodynamic diagram and the negative electrode tab end point thermodynamic diagram includes:
s61, performing non-local maximum suppression on the pixels in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
s62, selecting a pixel with a pixel value larger than a pixel threshold value according to the set pixel threshold value;
s63, for the negative plate endpoint thermodynamic diagram, if the number of the selected points is less than k-1, selecting the points with the largest pixel values in the remaining pixel points until the number of the selected pixel points reaches k-1;
s64, for the positive plate endpoint thermodynamic diagram, if the number of the selected pixel points is less than k, selecting the point with the largest pixel value in the rest pixel points until the number of the selected pixel points reaches k;
and k is the number of layers of the positive plates in the segmentation image. Referring to the cell structure diagram shown in fig. 1, in one cell, the number of negative plates is 1 less than that of positive plates, and therefore k-1 is the number of layers of the negative plates in the segmentation image.
In a preferred embodiment, in step S61, when the non-local maximum values are suppressed for the pixels in the positive electrode sheet end point thermodynamic diagram and the negative electrode sheet end point thermodynamic diagram, both the non-local maximum values are set to 0.
In steps S63 and S64, for the negative plate endpoint thermodynamic diagram, if the number of selected points is more than k-1, the retention is performed, which may be caused by missing detection when detecting the number of layers by using a neural network detection model. Due to the fact that non-local maximum suppression is conducted, only the maximum pixel value of each thermal point in the negative plate end point thermodynamic diagram is reserved. Similarly, for the positive plate endpoint thermodynamic diagram, if the number of selected points is more than k, the reservation is performed, and the missing detection may occur when the layer number is detected through the neural network detection model. Due to the fact that non-local maximum value suppression is conducted, only the maximum pixel value of each heat force point in the positive plate terminal thermodynamic diagram is reserved. By the screening method, the layer number of the positive plate and the negative plate can be further determined, and missing detection is prevented.
As a preferred embodiment, after all the positive plate end points and the negative plate end points are screened out, the obtained end points are further checked and screened. Firstly, judging whether the coordinate position of each endpoint has the problem of undersize distance, then judging whether the gray value of each endpoint is in a reasonable interval, if the gray value of each endpoint is not in the reasonable interval, obtaining the endpoint as an unreasonable point, and deleting the point.
The method comprises the following steps of screening positive plate end points and negative plate end points in a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram, and screening single-side plate end points in a single-side plate thermodynamic diagram. And cutting the image of the area where the single-surface patch is located after screening to obtain a check image, inputting the check image into the convolutional neural network model again to obtain a thermodynamic diagram, and screening again, so that further single-surface patch position checking is completed, and the false detection rate of the single-surface patch position is reduced. Wherein, the check-up image width includes 1 single face piece and 3 positive plates, and the check-up image aspect ratio is 1: 9.
in step S7, the abnormality detection based on the screened positive electrode tab end point and negative electrode tab end point includes:
and calculating the layer number, the coating value, the fall and the bending angle of the positive plate and the negative plate according to the position coordinates of the end points of the positive plate and the negative plate, and judging whether the battery cell is abnormal.
And if the fall and the bending angle of the pole piece exceed corresponding preset values, determining that the battery cell is abnormal.
Referring to fig. 9, in some embodiments, there is provided a cell abnormality detection apparatus of a laminated battery, including:
the acquisition module 101 is used for acquiring an image of the battery cell to be detected through X-Ray equipment;
the preprocessing module 102 is configured to preprocess the to-be-detected battery cell image and extract an image of an area of interest;
the layer number calculating module 103 is configured to input the region-of-interest image to a pre-established neural network detection model, and obtain the layer number of the positive plate and the negative plate in the region-of-interest image;
a segmentation module 104, configured to segment the region-of-interest image according to the number of layers to obtain a segmented image;
the feature identification module 105 is used for inputting the segmentation images into a pre-established convolutional neural network model to obtain a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram in each segmentation image;
the screening module 106 is used for screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
and the abnormality detection module 107 is used for performing abnormality detection on the screened positive plate end point and the screened negative plate end point.
Specifically, the preprocessing module 102 is further configured to perform normalization processing on the to-be-detected cell image to obtain an 8-bit grayscale image; extracting boundary points of the 8-bit gray level image according to a set gray level threshold value; and determining boundary points and two side boundary points of one end part of the positive plate and the negative plate in the 8-bit gray level image, and intercepting an area containing the boundary points and the two side boundary points of one end part of the positive plate and the negative plate as the region-of-interest image.
Further, the layer number calculating module 103 is further configured to identify a single-sided sheet of the battery cell through the neural network detection model, where the single-sided sheet is used to distinguish two or more battery cells.
The convolutional neural network model comprises a first convolutional layer, a first residual module, a first maximum pooling layer, a second residual module, a second maximum pooling layer, a second convolutional layer and a third convolutional layer.
Further, the feature identification module 105 is further configured to input the segmented image into the first convolution layer to perform 3 × 3 convolution with a step size of 2, so as to obtain a first feature map; inputting the first feature map into the first residual error module and the first maximum pooling layer, and respectively calculating to obtain a second feature map; inputting the second feature map into a second residual error module and a second maximum pooling layer to be calculated respectively, and then obtaining a third feature map; after the third feature map is subjected to upsampling, the third feature map is spliced with the second feature map and is input into the second convolution layer to be subjected to 1-by-1 convolution, and a convolution result is obtained; and after the convolution result is subjected to up-sampling, the convolution result is spliced with the first characteristic diagram and then input to the third convolution layer for 1-1 convolution, and a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram are obtained.
Wherein, establishing the convolutional neural network model comprises:
establishing an initial convolutional neural network model;
collecting a pole piece image sample, and processing the pole piece image sample to enable the pixel values of a positive pole piece endpoint and a negative pole piece endpoint in the pole piece sample image to be 1;
performing Gaussian filtering on the processed pole piece image sample to enable the pixel value far away from the positive pole piece end point and the negative pole piece end point in the pole piece image sample to be gradually reduced to 0;
and inputting the pole piece image sample subjected to Gaussian filtering into the initial convolutional neural network model for training, calculating the value of a loss function in the training, and obtaining the convolutional neural network model when the value of the loss function is smaller than a preset value.
Wherein the loss function is a mean square error loss function.
The pixel value of each pixel point in the positive plate endpoint thermodynamic diagram is the probability that the position is the positive plate endpoint, and the pixel value of each pixel point in the negative plate endpoint thermodynamic diagram is the probability that the position is the negative plate endpoint;
further, the screening module 106 is further configured to perform non-local maximum suppression on the pixels in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram; selecting pixel points with pixel values larger than the pixel threshold value according to the set pixel threshold value; for the negative plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k-1, selecting the pixels with the largest pixel values from the rest pixels until the number of the selected pixels reaches k-1; for the positive plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k, selecting the pixels with the largest pixel values in the rest pixels until the number of the selected pixels reaches k;
and k is the number of layers of the positive plates in the segmentation image.
Further, the abnormality detection module 107 is further configured to calculate the number of layers, the coating value, the fall and the bending angle of the positive plate and the negative plate according to the position coordinates of each of the positive plate end point and the negative plate end point, and determine whether the battery cell is abnormal.
Referring to fig. 10, in some embodiments, a cell abnormality detection system of a laminated battery is provided, which includes an X-Ray apparatus 3, a processor 1, and a storage device 2, where the storage device 2 stores a plurality of instructions, and the processor 1 is configured to read the instructions and execute the above-mentioned method.
Before the characteristic identification is performed on the image, the number of layers of the positive plate and the negative plate is calculated through the neural network model in addition to the binaryzation extraction boundary point, and then the image is divided into a plurality of smaller divided images as input data of the convolutional neural network model according to the number of layers, so that the identification accuracy is better when the complex detection condition is met, and the operation efficiency is improved; the pole piece characteristics are identified through a pre-trained convolutional neural network model to obtain a thermodynamic diagram so as to determine the end point position of the pole piece, and compared with the traditional methods such as setting a threshold value and the like, the method has better identification accuracy and greatly reduces the false detection rate; the method has the advantages that the single-surface plate is identified while the positive plate and the negative plate are identified, the multiple battery cores are distinguished through the single-surface plate, the abnormal detection of the battery is simplified into the abnormal detection of the battery cores, so that the method can carry out the abnormal detection on the battery containing the multiple battery cores, the endpoints of the positive plate and the negative plate are determined by establishing a convolutional neural network model with the output pixel values representing the endpoint probability, and the missing detection can be effectively avoided compared with the mode of directly detecting the pixel coordinates by adopting the model in the prior art.
The method and the system for detecting the abnormity of the battery core of the laminated battery provided by the embodiment are used for detecting the abnormity of the battery, determining good products and defective products, wherein the false detection rate is less than or equal to 2% in the detection, the detection time of a single battery core or battery is less than 1s, and the abnormity of the battery containing a plurality of battery cores can be detected.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention. It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (9)
1. A method for detecting the abnormity of a battery core of a laminated battery is characterized by comprising the following steps:
acquiring an image of the battery cell to be detected through X-Ray equipment;
preprocessing the to-be-detected battery cell image, and extracting an interested area image;
inputting the region-of-interest image into a pre-established neural network detection model to obtain the number of layers of the positive plate and the negative plate in the region-of-interest image;
segmenting the image of the region of interest according to the layer number to obtain a segmented image;
inputting the segmentation images into a pre-established convolutional neural network model to obtain positive plate end point thermodynamic diagrams and negative plate end point thermodynamic diagrams in each segmentation image;
screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
performing abnormity detection based on the screened positive plate end point and the screened negative plate end point;
the pixel value of each pixel point in the positive plate endpoint thermodynamic diagram is the probability that the position is the positive plate endpoint, and the pixel value of each pixel point in the negative plate endpoint thermodynamic diagram is the probability that the position is the negative plate endpoint;
screening the positive plate end points and the negative plate end points in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram, and the screening method comprises the following steps:
carrying out non-local maximum suppression on pixels in the positive plate end point thermodynamic diagram and the negative plate end point thermodynamic diagram;
selecting pixel points with pixel values larger than the pixel threshold value according to the set pixel threshold value;
for the negative plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k-1, selecting the pixels with the largest pixel values from the rest pixels until the number of the selected pixels reaches k-1;
for the positive plate endpoint thermodynamic diagram, if the number of the selected pixels is less than k, selecting the pixels with the largest pixel values in the rest pixels until the number of the selected pixels reaches k;
and k is the number of layers of the positive plates in the segmentation image.
2. The method according to claim 1, wherein the preprocessing is performed on the image of the to-be-detected battery cell to extract an image of a region of interest, and the method comprises the following steps:
carrying out normalization processing on the to-be-detected battery cell image to obtain an 8-bit gray level image;
extracting boundary points of the 8-bit gray level image according to a set gray level threshold value;
and determining boundary points and two side boundary points of one end part of the positive plate and the negative plate in the 8-bit gray level image, and intercepting an area containing the boundary points and the two side boundary points of one end part of the positive plate and the negative plate as the region-of-interest image.
3. The method according to claim 1, wherein the obtaining of the number of layers of the positive plate and the negative plate in the region of interest image further comprises:
and identifying a single-side sheet of the battery cell through the neural network detection model, wherein the single-side sheet is used for distinguishing two or more battery cells.
4. The method of claim 1, wherein the convolutional neural network model comprises a first convolutional layer, a first residual module, a first max pooling layer, a second residual module, a second max pooling layer, a second convolutional layer, and a third convolutional layer.
5. The method according to claim 4, wherein inputting the segmentation images into a pre-established convolutional neural network model to obtain a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram in each segmentation image comprises:
inputting the segmentation image into the first convolution layer to carry out 3-by-3 convolution with the step length of 2 to obtain a first feature map;
inputting the first feature map into the first residual error module and the first maximum pooling layer, and respectively calculating to obtain a second feature map;
inputting the second feature map into the second residual error module and a second maximum pooling layer, and respectively calculating to obtain a third feature map;
after the third feature map is subjected to upsampling, the third feature map is spliced with the second feature map and is input into the second convolution layer to be subjected to 1-by-1 convolution, and a convolution result is obtained;
and after the convolution result is subjected to up-sampling, the convolution result is spliced with the first characteristic diagram and then input to the third convolution layer for 1-1 convolution, and a positive plate end point thermodynamic diagram and a negative plate end point thermodynamic diagram are obtained.
6. The method of claim 1 or 4, wherein building a convolutional neural network model comprises:
establishing an initial convolutional neural network model;
collecting a pole piece image sample, and processing the pole piece image sample to enable the pixel values of a positive pole piece endpoint and a negative pole piece endpoint in the pole piece image sample to be 1;
performing Gaussian filtering on the processed pole piece image sample to enable the pixel values far away from the positive pole piece endpoint and the negative pole piece endpoint in the pole piece image sample to gradually decrease to 0;
and inputting the pole piece image sample subjected to Gaussian filtering into the initial convolutional neural network model for training, calculating the value of a loss function in the training, and obtaining the convolutional neural network model when the value of the loss function is smaller than a preset value.
7. The method of claim 6, wherein the loss function is a mean square error loss function.
8. The method according to claim 1, wherein the abnormality detection based on the screened positive plate end point and the screened negative plate end point comprises:
and calculating the layer number, the coating value, the fall and the bending angle of the positive plate and the negative plate according to the position coordinates of the end points of the positive plate and the negative plate, and judging whether the battery cell is abnormal.
9. A cell abnormality detection system for a laminated battery, comprising an X-Ray device, a processor, and a storage device, wherein the storage device stores a plurality of instructions, and the processor is configured to read the instructions and execute the method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210485397.2A CN114581446B (en) | 2022-05-06 | 2022-05-06 | Battery core abnormity detection method and system of laminated battery |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210485397.2A CN114581446B (en) | 2022-05-06 | 2022-05-06 | Battery core abnormity detection method and system of laminated battery |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114581446A CN114581446A (en) | 2022-06-03 |
| CN114581446B true CN114581446B (en) | 2022-08-09 |
Family
ID=81769393
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210485397.2A Active CN114581446B (en) | 2022-05-06 | 2022-05-06 | Battery core abnormity detection method and system of laminated battery |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114581446B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115511834A (en) * | 2022-09-28 | 2022-12-23 | 广东利元亨智能装备股份有限公司 | Battery cell alignment degree detection method, controller, detection system and storage medium |
| CN116309265A (en) * | 2022-11-30 | 2023-06-23 | 宁德时代新能源科技股份有限公司 | Method and device for detecting electrode plate of battery cell and electronic equipment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886971A (en) * | 2019-01-24 | 2019-06-14 | 西安交通大学 | A kind of image partition method and system based on convolutional neural networks |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109598721B (en) * | 2018-12-10 | 2021-08-31 | 广州市易鸿智能装备有限公司 | Defect detection method and device for battery pole piece, detection equipment and storage medium |
| CN109786492A (en) * | 2019-03-07 | 2019-05-21 | 苏州阿特斯阳光电力科技有限公司 | Photovoltaic module and its manufacturing method |
| CN111625998A (en) * | 2020-05-29 | 2020-09-04 | 华中科技大学 | Method for optimizing structure of laminated solar cell |
| CN212620580U (en) * | 2020-07-31 | 2021-02-26 | 蜂巢能源科技有限公司 | Battery lamination alignment degree detection device and lamination equipment |
| CN214898531U (en) * | 2020-11-20 | 2021-11-26 | 恒大新能源技术(深圳)有限公司 | Battery core lamination equipment and lamination battery core |
| CN112686915A (en) * | 2021-03-11 | 2021-04-20 | 聚时科技(江苏)有限公司 | Photovoltaic module picture cutting method based on full convolution neural network |
| CN113654493A (en) * | 2021-08-13 | 2021-11-16 | 苏州市比特优影像科技有限公司 | Quality detection method and system for laminated soft package lithium battery |
| CN113505865B (en) * | 2021-09-10 | 2021-12-07 | 浙江双元科技股份有限公司 | Sheet surface defect image recognition processing method based on convolutional neural network |
| CN113936004B (en) * | 2021-12-20 | 2022-03-18 | 浙江双元科技股份有限公司 | Lithium battery welding quality detection method, device and system |
-
2022
- 2022-05-06 CN CN202210485397.2A patent/CN114581446B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886971A (en) * | 2019-01-24 | 2019-06-14 | 西安交通大学 | A kind of image partition method and system based on convolutional neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114581446A (en) | 2022-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN114581446B (en) | Battery core abnormity detection method and system of laminated battery | |
| CN115511889A (en) | Method for detecting welding defects on surface of solar cell panel bracket | |
| CN112508090A (en) | External package defect detection method | |
| CN113205063A (en) | Visual identification and positioning method for defects of power transmission conductor | |
| CN116721107B (en) | Intelligent monitoring system for cable production quality | |
| CN114820625B (en) | Automobile top block defect detection method | |
| CN115861210B (en) | Transformer substation equipment abnormality detection method and system based on twin network | |
| CN111275040A (en) | Positioning method and device, electronic equipment and computer readable storage medium | |
| CN110619146A (en) | Polycrystalline silicon cell crack defect detection method based on structural similarity measurement | |
| CN117333489B (en) | Film damage detection device and detection system | |
| CN113012157A (en) | Visual detection method and system for equipment defects | |
| CN115170518A (en) | Cell detection method and system based on deep learning and machine vision | |
| CN115035061A (en) | Method for extracting cathode and anode points of battery cell electrode group and detecting alignment degree of cathode and anode sheets | |
| CN117635609A (en) | Visual inspection method for production quality of plastic products | |
| CN117372812A (en) | Intelligent identification method for battery CT image pole piece alignment degree based on network learning | |
| CN117078678B (en) | Waste silicon wafer shape detection method based on image recognition | |
| CN117541585A (en) | Method and device for detecting exposed foil defect of lithium battery pole piece | |
| CN116416268B (en) | Method and device for detecting edge position of lithium battery pole piece based on recursion dichotomy | |
| CN116363136B (en) | On-line screening method and system for automatic production of motor vehicle parts | |
| CN115861220A (en) | Cold-rolled strip steel surface defect detection method and system based on improved SSD algorithm | |
| CN111429411B (en) | X-ray defect image sample generation method for carbon fiber composite core wire | |
| CN115439319A (en) | Exposed detection method for electric slide wire protection device | |
| CN113506290A (en) | Method and device for detecting defects of line insulator | |
| CN111523583A (en) | Method for automatically identifying and classifying equipment nameplate photos by using unmanned aerial vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |