1. A method for screening defective batteries based on machine learning is characterized by comprising the following steps

Acquiring parameter data of a battery to be tested in a high-temperature state, wherein the parameter data comprises a first open-circuit voltage, a first alternating current internal resistance and a high-temperature self-discharge rate;

the method comprises the steps of initially screening abnormal self-discharge batteries at high temperature, constructing a first coordinate model based on the obtained first open-circuit voltage, first alternating current internal resistance and high-temperature self-discharge rate, carrying out outlier detection on data in the first coordinate model through a machine learning algorithm, judging the batteries corresponding to the outlier coordinates as suspected defective batteries, and judging the batteries corresponding to the non-outlier coordinates as normal batteries;

performing self-discharge at high temperature, performing interpolation on the first coordinate model to obtain a first reference curved surface according to each first open-circuit voltage and the lowest high-temperature self-discharge rate under first alternating internal resistance of a normal battery in the batteries to be tested, and changing the suspected defective battery corresponding to a coordinate point of which the high-temperature self-discharge rate is under the first reference curved surface into the normal battery, wherein the first open-circuit voltage is between the minimum and maximum open-circuit voltage values of the normal battery, and the first alternating internal resistance is between the minimum and maximum alternating internal resistance values of the normal battery;

screening the batteries with abnormal capacity, acquiring the capacity of the batteries, acquiring parameter data of the batteries to be detected at normal temperature, including third open circuit voltage and second alternating current internal resistance, constructing a second coordinate model based on the acquired third open circuit voltage, the second alternating current internal resistance and the battery capacity data, performing outlier detection on data in the second coordinate model through a machine learning algorithm, judging the batteries corresponding to the outlier coordinates as defective batteries with abnormal capacity, and judging the batteries corresponding to non-outliers as normal batteries;

acquiring parameter data of a battery to be tested in a normal temperature state, wherein the parameter data comprises a fourth circuit voltage, a third alternating current internal resistance and a normal temperature self-discharge rate;

performing primary screening on the abnormal self-discharge battery at the normal temperature, constructing a third coordinate model based on the obtained fourth circuit voltage, the third alternating current internal resistance and the normal-temperature self-discharge rate, performing outlier detection on data in the third coordinate model through a machine learning algorithm, judging the battery corresponding to the outlier coordinate as a suspected defect battery, and changing the battery corresponding to the non-outlier coordinate into a normal battery;

performing self-discharge abnormal battery rescreening at normal temperature, interpolating on a third coordinate model to obtain a second reference curved surface according to each fourth circuit voltage and the lowest normal-temperature self-discharge rate under a third alternating current internal resistance of a normal battery to be tested, and changing the suspected defective battery corresponding to a coordinate point of which the normal-temperature self-discharge rate is under the second reference curved surface into a normal battery, wherein the fourth circuit voltage is between the minimum and maximum open circuit voltage values of the normal battery, and the third alternating current internal resistance is between the minimum and maximum alternating current internal resistance values of the normal battery;

screening the batteries with abnormal internal resistance, acquiring a fifth open-circuit voltage and direct-current internal resistance of the battery to be detected in a normal temperature state, constructing a fourth coordinate model based on the fifth open-circuit voltage and the direct-current internal resistance, performing outlier detection on data in the fourth coordinate model through a machine learning algorithm, judging the battery corresponding to the coordinates corresponding to the outlier as a defective battery with abnormal internal resistance, and judging the battery corresponding to the coordinates of the non-outlier as a normal battery;

the high-temperature state temperature is 40-50 ℃, and the normal-temperature state temperature is 20-30 ℃.

2. The method for screening defective batteries based on machine learning according to claim 1, wherein the method comprises the following steps of outputting the results of the screened defective batteries after high-temperature self-discharge abnormality detection, capacity abnormality detection, normal-temperature self-discharge abnormality detection and internal resistance abnormality self-discharge detection, and finally screening the screened normal batteries:

the method comprises the steps of obtaining the highest value of the internal resistance, the high-temperature self-discharge rate and the normal-temperature self-discharge rate of a certain number of screened normal batteries and the lowest value of the capacity, presetting the highest value and the range value of the lowest value according to the production line yield requirement, and judging that the battery needing to be subjected to final screening is a defective battery if the internal resistance, the high-temperature self-discharge rate and the normal-temperature self-discharge rate of the battery needing to be subjected to final screening fall into the range of the highest value and the value of the capacity of the battery needing to be subjected to final screening fall into the range of the lowest value.

3. The method for screening the defective battery based on the machine learning as claimed in claim 2, wherein in the screening and detecting process, a sliding mechanism is adopted to obtain the data of the battery to be tested, and the sliding mechanism is: in the process of battery production line production, battery sample data is sequentially acquired to form a sample data group with n adjacent battery sample data, the battery sample data acquired first in the sample data group is replaced to form a new sample data group after new battery sample data is acquired, and during detection and screening, data in the sample data group is detected each time, and the latest data in the sample data group is subjected to result judgment.

4. The method for screening defective batteries based on machine learning according to claim 3, wherein the first open-circuit voltage is measured by standing the batteries in an environment of 40-50 ℃ for 10-14h, the second open-circuit voltage is measured after standing for 3-5 days at the temperature, and the self-discharge rate is the ratio of the difference value between the first open-circuit voltage and the second open-circuit voltage to the standing time.

5. The method for screening the defective battery based on the machine learning as claimed in claim 4, wherein when screening the battery with abnormal capacity, the battery capacity is obtained by fully discharging the battery, and the open-circuit voltage is measured by standing for 4-8h to obtain the third open-circuit voltage;

and standing for 4-8h at the normal temperature after the third open circuit voltage is measured, and measuring the alternating current internal resistance to obtain the second alternating current internal resistance.

6. The defective battery screening method based on machine learning of claim 4, wherein the fourth circuit voltage and the third alternating current internal resistance are obtained by recharging the batteries with the divided capacity and standing for 4-8h to measure the open circuit voltage and the alternating current internal resistance at normal temperature;

and after the fourth open-circuit voltage is obtained, standing for 7-14 days at the normal temperature state, and measuring the open-circuit voltage again to obtain a fifth open-circuit voltage, wherein the normal-temperature self-discharge rate is the ratio of the pressure difference between the fifth open-circuit voltage and the fourth open-circuit voltage to the standing time.

7. The defective battery screening method based on machine learning of claim 1, wherein after the coordinate model is constructed, the machine learning algorithm for outlier detection comprises isolated forest, K-means clustering or local outlier factors.

8. The defective battery screening method based on machine learning of claim 1, wherein a coordinate model is built according to data collected from a battery sample;

the first coordinate model and the third coordinate model use the obtained open-circuit voltage as an x coordinate, the alternating current internal resistance as a y coordinate, and the self-discharge rate as a z coordinate;

the second coordinate model takes open-circuit voltage as an x coordinate, alternating-current internal resistance as a y coordinate, and battery capacity as a z coordinate;

the fourth coordinate model takes open-circuit voltage as an abscissa and direct-current resistance as an ordinate.

9. A defective battery screening device is characterized by comprising a detection terminal, a data acquisition device and a battery production line device;

the inspection terminal stores a computer program for executing the method of any one of claims 1 to 8 for inspecting defective batteries and normal batteries of a battery sample produced in the battery production line apparatus and outputting the results;

the data acquisition device is used for acquiring parameter data of a battery sample to be detected in the battery production line device and transmitting the data to the detection terminal;

the battery production line device is used for producing batteries and providing a temperature environment for battery detection.

10. A computer storage medium storing a computer program, characterized in that the computer program, when being executed by a processor, carries out the steps of the method according to any one of claims 1-8.

Background

In order to deal with the problems of carbon emission, environmental pollution, energy crisis and the like, new energy industries, particularly the battery industry, are vigorously developed worldwide. In the mass production process of lithium ion batteries, after the formation (first charge activation) of the batteries is completed, a self-discharge rate test (commonly called a K value test), a capacity test and an internal resistance test (commonly called a DCR test) are generally performed to screen out defective bad batteries and avoid safety accidents caused by the fact that the batteries flow into the market.

The method has the following problems that firstly, the setting of the threshold value can only depend on experience, sieving and sieve leakage conditions exist, the method cannot be well adapted to batteries in different production batches, not only can the yield be low, but also the quality of the batteries is poor, and even the safety problem of the batteries at the market end can be caused; secondly, each detection item has no correlation, and because of different thresholds, misjudgment can be caused to the battery with marginal detection data; and the method often needs large-batch battery accumulation, so that the inventory pressure of enterprises is high, and the product circulation efficiency is low.

Disclosure of Invention

The invention provides a defective battery screening method, equipment and a medium based on machine learning, wherein in the production process of batteries, a machine learning algorithm is utilized to carry out self-discharge rate, capacity and internal resistance tests on a large batch of batteries, outliers of detection data are utilized to screen and judge batteries to be tested, a re-screening mechanism is arranged to retrieve batteries with normal outliers to avoid screening, and finally a final screening mechanism is arranged to avoid screen leakage by judging whether batteries with extreme values of various data appear in each screening process.

The invention provides a defective battery screening method, equipment and a medium based on machine learning, and the specific scheme is as follows:

acquiring parameter data of a battery to be tested in a high-temperature state, wherein the parameter data comprises a first open-circuit voltage, a first alternating current internal resistance and a high-temperature self-discharge rate;

the method comprises the steps of initially screening abnormal self-discharge batteries at high temperature, constructing a first coordinate model based on the obtained first open-circuit voltage, first alternating current internal resistance and high-temperature self-discharge rate, carrying out outlier detection on data in the first coordinate model through a machine learning algorithm, judging the batteries corresponding to the outlier coordinates as suspected defective batteries, and judging the batteries corresponding to the non-outlier coordinates as normal batteries;

performing self-discharge at high temperature, performing interpolation on the first coordinate model to obtain a first reference curved surface according to each first open-circuit voltage and the lowest high-temperature self-discharge rate under first alternating internal resistance of a normal battery in the batteries to be tested, and changing the suspected defective battery corresponding to a coordinate point of which the high-temperature self-discharge rate is under the first reference curved surface into the normal battery, wherein the first open-circuit voltage is between the minimum and maximum open-circuit voltage values of the normal battery, and the first alternating internal resistance is between the minimum and maximum alternating internal resistance values of the normal battery;

screening the batteries with abnormal capacity, acquiring the capacity of the batteries, acquiring parameter data of the batteries to be detected at normal temperature, including third open circuit voltage and second alternating current internal resistance, constructing a second coordinate model based on the acquired third open circuit voltage, the second alternating current internal resistance and the battery capacity data, performing outlier detection on data in the second coordinate model through a machine learning algorithm, judging the batteries corresponding to the outlier coordinates as defective batteries with abnormal capacity, and judging the batteries corresponding to non-outliers as normal batteries;

acquiring parameter data of a battery to be tested in a normal temperature state, wherein the parameter data comprises a fourth circuit voltage, a third alternating current internal resistance and a normal temperature self-discharge rate;

performing primary screening on the abnormal self-discharge battery at the normal temperature, constructing a third coordinate model based on the obtained fourth circuit voltage, the third alternating current internal resistance and the normal-temperature self-discharge rate, performing outlier detection on data in the third coordinate model through a machine learning algorithm, judging the battery corresponding to the outlier coordinate as a suspected defect battery, and changing the battery corresponding to the non-outlier coordinate into a normal battery;

performing self-discharge abnormal battery rescreening at normal temperature, interpolating on a third coordinate model to obtain a second reference curved surface according to each fourth circuit voltage of a normal battery in the battery to be tested and the lowest normal-temperature self-discharge rate under first third alternating current internal resistance, changing the suspected defective battery corresponding to a coordinate point of which the normal-temperature self-discharge rate is under the second reference curved surface into a normal battery, wherein the fourth circuit voltage is between the minimum and maximum open-circuit voltage values of the normal battery, and the third alternating current internal resistance is between the minimum and maximum alternating current internal resistance values of the normal battery;

screening the batteries with abnormal internal resistance, acquiring a fifth open-circuit voltage and direct-current internal resistance of the battery to be detected in a normal temperature state, constructing a fourth coordinate model based on the fifth open-circuit voltage and the direct-current internal resistance, performing outlier detection on data in the fourth coordinate model through a machine learning algorithm, judging the battery corresponding to the coordinates corresponding to the outlier as a defective battery with abnormal internal resistance, and judging the battery corresponding to the coordinates of the non-outlier as a normal battery.

In the screening and detecting process, a sliding mechanism is adopted to obtain the data of the battery to be detected, and the sliding mechanism is as follows: in the process of battery production line production, battery sample data is sequentially acquired to form a sample data group with n adjacent battery sample data, the battery sample data acquired first in the sample data group is replaced to form a new sample data group after new battery sample data is acquired, and during detection and screening, data in the sample data group is detected each time, and the latest data in the sample data group is subjected to result judgment.

Further, the method comprises the following specific steps of outputting the result of the defective battery obtained after screening through high-temperature self-discharge abnormality detection, capacity abnormality detection, normal-temperature self-discharge abnormality detection and internal resistance abnormality self-discharge detection, and finally screening the obtained result of the defective battery, wherein the specific process comprises the following steps:

and acquiring the internal resistance, the high-temperature self-discharge rate and the normal-temperature self-discharge rate of the normal battery and the lowest value of the capacity in the screening result, judging the normal battery close to the preset difference range of the highest value and the lowest value as a defective battery, and outputting the defective battery.

Further, the first open-circuit voltage is obtained by standing the battery for 10-14h in an environment of 40-50 ℃, the second open-circuit voltage is obtained by standing the battery for 3-5 days at the temperature, and the self-discharge rate is the ratio of the difference value between the first open-circuit voltage and the second open-circuit voltage to the standing time.

Further, the state temperature of the normal temperature is 20-30 ℃.

Further, when screening the batteries with abnormal capacity, fully discharging the batteries to obtain the capacity of the batteries, standing for 4-8h for measuring open-circuit voltage to obtain third open-circuit voltage;

and standing for 4-8h at the normal temperature after the third open circuit voltage is measured, and measuring the alternating current internal resistance to obtain the second alternating current internal resistance.

Further, at the normal temperature, the batteries with the capacity grading completed are charged again and stand still for 4-8 hours to measure the open-circuit voltage and the alternating current internal resistance, so that the fourth open-circuit voltage and the third alternating current internal resistance are obtained;

and after the fourth open-circuit voltage is obtained, standing for 7-14 days at the normal temperature state, and measuring the open-circuit voltage again to obtain a fifth open-circuit voltage, wherein the normal-temperature self-discharge rate is the ratio of the pressure difference between the fifth open-circuit voltage and the fourth open-circuit voltage to the standing time.

Further, after the coordinate model is constructed, the machine learning algorithm for detecting the outliers comprises isolated forests, K-means clustering or local outlier factors.

Further, a coordinate model is built according to data collected by the battery sample;

the first coordinate model and the third coordinate model use the obtained open-circuit voltage as an x coordinate, the alternating current internal resistance as a y coordinate, and the self-discharge rate as a z coordinate;

the second coordinate model takes open-circuit voltage as an x coordinate, alternating-current internal resistance as a y coordinate, and battery capacity as a z coordinate;

the fourth coordinate model takes open-circuit voltage as an abscissa and direct-current resistance as an ordinate.

The invention provides defective battery screening equipment which comprises a detection terminal, a data acquisition device and a battery production line device, wherein the detection terminal is connected with the data acquisition device;

the detection terminal is used for detecting defective batteries and normal batteries of the battery samples produced in the battery production line device and outputting results;

the data acquisition device is used for acquiring parameter data of a battery sample to be detected in the battery production line device and transmitting the data to the detection terminal;

the battery production line device is used for producing batteries and providing a temperature environment for battery detection.

The present invention provides a computer storage medium storing a computer program which, when executed by a processor, implements the steps of the above-described defective battery screening method based on machine learning.

The invention has the following beneficial effects:

1. the battery test result is analyzed and diagnosed based on the machine learning algorithm, the phenomenon that a threshold value is set only by traditional experience and is missed to be sieved or sieved due to inaccurate threshold value setting is avoided, the battery production line yield can be improved while the battery quality is guaranteed, and the battery test result analysis method is suitable for being used when batteries are produced in large batches by battery production enterprises.

2. The batch processing battery data is used for screening and judging the batteries, is suitable for the production of battery production lines, does not cause the accumulation of a large number of batteries, reduces the inventory pressure of enterprises, and improves the product circulation efficiency.

3. A re-screening mechanism is arranged in the self-discharge detection, a reference curved surface is set through interpolation, judgment is carried out again according to the reference curved surface, a battery sample which is judged wrongly in the primary screening is changed into a normal battery, misjudgment is avoided, and the accuracy of battery screening detection is improved.

4. And finally, a final screening mechanism is set for battery detection, results of all screening detections are correlated, all detection data in a normal battery are close to a battery standard position defect battery of the extreme value of the whole battery sample, and the condition that the defective battery of critical data is missed to be screened because all screening is uncorrelated is avoided.

Drawings

FIG. 1 is a schematic flow diagram of the process of the present invention;

FIG. 2 is a schematic diagram of the structure of the computer device of the present invention.

Detailed Description

In the following description, technical solutions in the embodiments of the present invention are clearly and completely described, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Example 1

Embodiment 1 of the present invention provides a defective battery screening method based on machine learning.

Firstly, setting batch detection samples, and selecting 15000 batteries on a battery production line as to-be-detected battery samples in the embodiment; in this embodiment, the detection sequence includes high-temperature self-discharge screening detection, capacity screening detection, normal-temperature self-discharge screening detection, and internal resistance screening detection, and the detection sequence can be adjusted according to actual conditions.

In the production process of a battery assembly line, after data of n batteries are obtained, screening detection is carried out by adopting a sliding mechanism, wherein the sliding mechanism is as follows: in the process of battery production line production, sequentially acquiring battery sample data to form a sample data group with n adjacent battery sample data, replacing the battery sample data acquired first in the sample data group to form a new sample data group after new battery sample data is obtained, detecting the data in the sample data group each time during detection and screening, and judging the result of the latest data in the sample data group; in the battery production line production process, after sample data of adjacent n batteries is obtained, outlier detection is carried out according to the sample data of the 1 st to n batteries, and if the point corresponding to the nth battery is the outlier, the battery is judged to be abnormal; and after the n +1 th battery sample data is obtained, performing outlier detection on the sample data of the n +1 th battery by using the previous 2 nd to n +1 th battery data, and if the point corresponding to the n +1 th battery is the outlier, judging the battery to be an abnormal battery, and so on. The number of battery samples for each outlier detection is n, but the battery samples for each algorithm processing are sliding.

The self-discharge amount of the normal battery is different from that of the defective battery, the voltage data of the battery is measured after the battery is placed for a period of time at a certain temperature and is used as the parameter data of the sample, and the temperature and the time can be correspondingly adjusted according to the actual situation.

Preferably, as shown in fig. 1, the method comprises the following steps:

s1: acquiring the open-circuit voltage, the alternating current internal resistance and the self-discharge rate of a battery to be detected;

after the battery is completed, placing the battery to be tested in an environment at 50 ℃, standing for 12h, measuring the voltage of the battery, recording the voltage obtained at the moment as a first open-circuit voltage, measuring the alternating current internal resistance of the battery as a first alternating current internal resistance, standing for 4 days at the current temperature state, measuring the voltage of the battery and recording the voltage as a second open-circuit voltage, calculating the first open-circuit voltage and the second open-circuit voltage to obtain a differential pressure, and dividing the differential pressure by the standing time at the high temperature state to obtain the high-temperature self-discharge rate of the battery to be tested.

S2: primary screening of the battery with abnormal self-discharge at high temperature: after first open-circuit voltage, first alternating current internal resistance and high-temperature self-discharge rate of the 15000 batteries of the batch are obtained in the production process of a battery assembly line, the current battery is taken as a sample, the first open-circuit voltage, the first alternating current internal resistance and the high-temperature self-discharge rate of the 15000 batteries form 15000 lines of 3 lines of data, the first open-circuit voltage is taken as an x coordinate, the first alternating current internal resistance is taken as a y coordinate, the high-temperature self-discharge rate is taken as a z coordinate to construct a rectangular coordinate model, the battery sample to be detected is mapped in the rectangular coordinate model, one point represents one battery, then the 15000 points are subjected to outlier detection by using an isolated forest algorithm or a K-mean clustering or local outlier factor, the battery corresponding to the outlier is a suspected defective battery, and the battery corresponding to the non-outlier is a normal battery. And if the point corresponding to the current battery (15000 th battery sample) is one of the outliers, determining that the current battery is a suspected defective battery, and if the point corresponding to the current battery is one of the non-outliers, determining that the current battery is a normal battery.

S3: re-screening the abnormal self-discharge battery at high temperature: obtaining the lowest high-temperature self-discharge rate under each first open-circuit voltage and each first alternating internal resistance in the normal battery obtained by the preliminary screening in the step S2, interpolating a reference curved surface, that is, based on the coordinate axis of the self-discharge rate in the coordinate model, using the battery coordinate points close to the xoz plane and the yoz plane as boundary points, using the plane formed by the boundary points and the z axis as a boundary surface, connecting all the battery sample points along the z axis in the boundary surface by curved surfaces, constructing a lowest high-temperature self-discharge rate curved surface, and further performing re-screening judgment, if the first open-circuit voltage of the suspected defective battery in the preliminary screening is between the minimum first open-circuit voltage value and the maximum first open-circuit voltage value of the normal battery, the first alternating internal resistance is between the minimum first alternating internal resistance value and the maximum first alternating internal resistance value of the normal battery, and the self-discharge rate is below the reference curved surface, if the self-discharge rate is lower than the lowest high-temperature self-discharge rate under the first open-circuit voltage and the first alternating current internal resistance of the normal battery, the suspected defective battery is judged as the normal battery. Since some of the cells in S2 belong to discrete points, they are considered normal cells because their self-discharge rates are low. And (3) carrying out the processing of steps S1-S3 on the subsequent batteries of the battery production line, namely judging the first batch of sample batteries according to the data of the 1 st to 15000 th batteries on the production line, then judging the 15001 th battery by using the data of the 2 nd to 15001 th batteries, and so on, thereby realizing sliding screening and adapting to the production process of the production line batteries.

S4: screening of the battery with abnormal capacity: fully charging the battery to be tested to obtain the battery capacity, standing the battery for 6 hours to remove the polarization influence, then detecting the voltage and the alternating current internal resistance of the batteries again, recording the obtained voltage and the alternating current internal resistance as third open circuit voltage and second alternating current internal resistance, forming 15000 lines of 3 columns of data by the third open circuit voltage, the second alternating current internal resistance and the battery capacity of the 15000 batteries, constructing a rectangular coordinate model by taking the third open circuit voltage as an x coordinate, the second alternating current internal resistance as a y coordinate and the battery capacity as a z coordinate, mapping a battery sample to be tested in the coordinate model, one point represents a battery, then, machine learning algorithms such as isolated forests, K-means clustering or local outlier factors and the like are utilized to carry out outlier detection on the battery sample points, batteries corresponding to the outliers are marked as defective batteries with abnormal capacity, and batteries corresponding to non-outliers are marked as normal batteries.

S5: acquiring open-circuit voltage, alternating current internal resistance and self-discharge rate of the batch of batteries to be tested at normal temperature;

in the production process of the battery assembly line, after the capacity grading test of the battery is completed, standing the battery to be tested in an environment at 20 ℃ for 6 hours, measuring the voltage of the battery as a fourth open-circuit voltage, measuring the alternating-current internal resistance of the battery as a third alternating-current internal resistance, standing the battery to be tested in the normal-temperature state for 12 days, measuring the voltage of the battery again as a fifth open-circuit voltage, calculating the pressure difference between the fourth open-circuit voltage and the fifth open-circuit voltage, and dividing the pressure difference by the standing time at normal temperature to obtain the normal-temperature self-discharge rate of the battery.

S5: primary screening of the self-discharge abnormal battery at normal temperature: and forming 15000 lines of 3-column data according to the obtained fourth circuit voltage, the third alternating current internal resistance and the normal-temperature self-discharge rate of the 15000 batteries to be detected, constructing a rectangular coordinate model by taking the fourth circuit voltage as an x coordinate, the third alternating current internal resistance as an x coordinate and the normal-temperature self-discharge rate as a z coordinate, mapping a battery sample to be detected in the coordinate model, wherein one point represents one battery, then performing outlier detection on the 15000 points by using an isolated forest algorithm, marking the batteries corresponding to the outliers as defective batteries with abnormal capacity, and marking the batteries corresponding to the non-outliers as normal batteries.

S6: re-screening the abnormal self-discharge battery at normal temperature:

obtaining the lowest normal temperature self-discharge rate under each fourth circuit voltage and the third alternating current internal resistance in the normal battery obtained by the preliminary screening in the S5, interpolating a reference curved surface, namely, based on the self-discharge rate coordinate axis in the coordinate model, taking the battery coordinate points close to xoz plane and yoz plane as boundary points, taking the plane formed by the boundary points and the z axis as the boundary surface, connecting all the battery sample points which are minimum along the z axis in the boundary surface to construct a lowest normal temperature self-discharge rate curved surface, further performing re-screening judgment, if the fourth circuit voltage of the suspected defective battery in the preliminary screening is between the minimum fourth circuit voltage value and the maximum fourth circuit voltage value of the normal battery, the third alternating current internal resistance is between the minimum third alternating current internal resistance value and the maximum third alternating current internal resistance value of the normal battery, and the normal temperature self-discharge rate is below the reference curved surface, namely the normal temperature self-discharge rate is lower than the fourth circuit voltage of the normal battery and the lowest high temperature self-discharge rate under the third alternating current internal resistance, and the suspected defective battery is judged as the normal battery. Similarly, a sliding mechanism is also adopted in the self-discharge abnormity detection process at normal temperature, and subsequent batteries of the battery production line are processed in steps S5-S6, namely, a first batch of sample batteries are judged according to data of 1-15000 th batteries on the production line, then the 15001 th batteries are judged according to data of 2-15001 th batteries, and the like, so that sliding screening is realized, and the method is suitable for the production process of the production line batteries.

S7: screening of batteries with abnormal internal resistance: after the fifth open-circuit voltage of the battery to be detected is obtained, direct-current internal resistance detection is carried out on the battery to be detected to obtain direct-current internal resistance, a rectangular coordinate model is constructed by taking the fifth open-circuit voltage as an abscissa and taking the internal resistance as an ordinate, one point represents one battery, then outlier detection is carried out on a sample point of the battery to be detected by utilizing machine learning algorithms such as isolated forests, K-means clustering or local outlier factors, the battery corresponding to the outlier is a defective battery with abnormal direct-current internal resistance, and the battery corresponding to the non-outlier is a normal battery.

S8: and (4) final screening of defective batteries: acquiring the highest values of the internal resistance, the high-temperature self-discharge rate and the normal-temperature self-discharge rate of the normal battery and the lowest value of the capacity, which are calibrated in the screening result of the step, and judging the normal battery close to the range of the preset difference value between the highest value and the lowest value as a defective battery; and (4) judging whether the internal resistance, the high-temperature self-discharge rate and the normal-temperature self-discharge rate of the normal battery obtained after the screening in the steps S1-S7 are close to the highest value and the capacity is close to the lowest value in the normal batteries in the other battery samples, if the internal resistance, the high-temperature self-discharge rate and the low-temperature self-discharge rate of the battery sample marked as the normal battery are close to the highest value and the capacity is close to the lowest value in the normal battery sample, marking the battery sample as a defective battery, and outputting the result.

Example 2

Embodiment 2 of the present invention provides a defective battery screening apparatus, as shown in fig. 2, including a detection terminal, a data acquisition device, and a battery production line device;

the detection terminal is used for detecting defective batteries and normal batteries of the battery samples produced in the battery production line device and outputting results;

the data acquisition device is used for acquiring parameter data of a battery sample to be detected in the battery production line device and transmitting the data to the detection terminal;

the battery production line device is used for producing batteries and providing a temperature environment for battery detection.

Example 3

Embodiment 3 of the present invention provides a computer storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described defective battery screening method based on machine learning.

The invention is not limited to the foregoing embodiments. The invention extends to any novel feature or any novel combination of features disclosed in this specification and any novel method or process steps or any novel combination of features disclosed.