The 2nd "Mobile Cloud Cup" Computing Power Network Application Innovation Competition
is in full swing
Do you still have a lot of question marks about the competition questions on each track?
So, follow us
Unlock the new posture of preparing for the game~
Today's main
Industry Track - Interpretation of Industrial Industry Application Sub-track
1
Background
Lithium-ion batteries have the advantages of high energy density, long cycle life, and low self-discharge rate. Tab welding is the key and difficult point in the entire lithium battery production process. At present, the commonly used tab welding is the ultrasonic method. Through the high-frequency vibration of the welding head, instantaneous high heat is generated, so that the tab and the cell foil produce intermolecular thermal fusion, so that there is a welding strength relationship between the two; if welding Insufficient strength will lead to a large internal resistance of the lithium battery, which will lead to the scrapping of the battery cell; if it is in a completely welded state, the small flow area will cause a large amount of heat to be generated during charging and discharging, resulting in safety risks such as the lithium battery catching fire. The detection of pole piece defects by machine vision has the advantages of no contact, no damage, and 100% detection, which is specifically reflected in: high accuracy, can accurately identify coating scratches, pole piece damage, and asymmetrical welding of tabs, etc., accurately The detection rate is much higher than that of human eyes; and it excludes the influence of human factors such as the subjective will, emotion, and visual fatigue of the test personnel on the battery test results, and has high reliability, fast speed, high efficiency and low cost.
2
Introduction
This track is open to developers around the world, exploring diversified computing power network application scenarios, and encouraging participants to develop and form rich computing power network innovative applications based on the capabilities of mobile cloud products.
This track takes "Machine Vision Inspection of Lithium Battery Tab Welding Quality" as the preliminary competition topic. It aims to use artificial intelligence technology to develop a set of lithium battery tab welding quality inspection algorithms, and use computer vision to replace human eyes for various quality, Safety and integrity inspection work.
The dataset introduces:
The data set is the welding pictures of lithium battery tabs marked by classification, and the classification includes "jssx metal debris (1), OK qualified (2), lj edge debris (3), ong other NG (4), tfhd carbon powder black spots (5)"5 categories, the pictures include training set and test set, and the training set and test set both cover the above five types of classification data. The training set data is open to the contestants, and the test set is used as reserved data for the test and evaluation of the model.
Note: For details of image examples of qualified samples and defective samples, please refer to the data download.
Specific submission material requirements:
1) Algorithm model;
2) detective script;
A function introduction. The predictor module is an inference script, which aims to load the trained model through this script, and obtain the analysis results of the samples to be analyzed in the test set according to the incoming test set data samples. The organizer will call the reasoning script program to obtain the specific experimental results and running time, and calculate the results through the evaluation script to obtain the evaluation score.
B input and output. The predictor is set as a class, and the contestant needs to specify the model saving location and load the model during initialization. When using the function predict in the class to perform inference operations, you need to set the incoming parameters (single image data path); the return value of the function should be the category number corresponding to the image in the test data set (jssx metal debris (1), OK qualified (2) , lj edge debris (3), ong other NG (4), tfhd toner black spots (5)).
Predictor的伪码说明如下：
import...
class Model(): # 主类
def __init__(self,p_path):
self.__model = None
# 使用按照压缩包路径下的相对路径，
# 如，解压后的文件夹名称为model,模型存放于model下的pysrc/model.pkl路径，则路径为'model/pysrc/model.pkl'
try:
# -------------------------如下位置可修改(加载模型)------------------------
self.load_model(p_path+'/model/pysrc/model.pkl')
# ------------------------- END ------------------------
except Exception as e:
...
return None
def predict(self,path_to_img:str): # 入参：单个图片的路径，如’aaa.jpg’
# Read in image
try:
# -------------------------如下位置可修改(读取图片)------------------------
x = Image.open(path_to_img)
# ------------------------- END ------------------------
except Exception as e:
...
# predict & score
try:
# -------------------------如下位置可修改(推理过程)------------------------
prediction = self.__model(x)
# ------------------------- END ------------------------
except Exception as e:
...
return prediction # 返回结果：整数，与label对应关系为：（jssx金属碎屑(1)、OK合格(2)、lj边缘碎屑（3）、ong其他NG（4）、tfhd碳粉黑点(5)）
def load_model(self,path_to_model:str):
# Load model
# -------------------------如下位置可修改（加载模型）------------------------
with open(path_to_model,'rb') as fp:
paras = pickle.load(fp)
self.__model = lambda x : random.randint(paras[0],paras[1])
# ------------------------- END ------------------------
return 0The pseudocode for calling the Predictor class is as follows:
if __name__ == '__main__':
M = Model('./') # 初始化模型
print(M.predict('2023_zyy_35.jpg')) # 调用推理过程
# 输出：1-5之间的整数
注：上传作品文件夹目录结构如下（样例代码供下载，见资料下载）：
├── model 用户上传目录
│ ├── __init__.py 包函数
│ ├── predictor.py 预测函数
│ └── pysrc 模型路径
│ ├── model.pkl 模型（格式可按照模型变更）The upload and related paths are subject to comments.
3
Competition arrangement
The competition is divided into five stages: registration, preliminaries, semi-finals, track finals, and finals. 15 teams advance to the preliminary rounds, 10 teams advance to the semi-finals, and the top 3 teams in the track finals are recommended to be selected for the national finals. Road Excellence Award. The teams that have been promoted in each review process should follow the unified arrangement of the competition to participate in the next round of competition review. If the team fails to participate due to personal reasons, it will be deemed as giving up the promotion quota for the next round, and the promotion quota (including bonus) will be postponed according to the ranking of this stage.
take a look
are you ready
full sprint
Fight for your dreams!