Full-automatic battery cell test board and test process optimization method thereof

Abstract

The invention discloses a full-automatic battery cell test board and a test flow optimization method thereof, and relates to the technical field of automatic test.The method comprises the steps that a battery cell test parameter database of each power battery type is established, and the local fault trend of the battery cell of each power battery type is evaluated; screening the cell test parameters of the same power battery type in the cell test parameter database to perform deviation state analysis, and evaluating the global fault trend of the cells of the same power battery type; according to the similarity between the cell local fault trend of each power battery type and the cell global fault trend of the same power battery type, determining cell global fault attention factors of each power battery type, and compensating for the cell local fault trend of each power battery type. And generating the local fault risk probability of the battery cell of each power battery type. The method has the advantages that the diagnosis accuracy is improved, various battery systems are compatible, and the battery testing efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic test technology, in particular to a full-automatic battery cell test platform and a test process optimization method thereof. BACKGROUND

[0002] The single-parameter threshold detection mode used in the existing power battery test leads to missed detection of multi-parameter coupling faults, lacks group data collaborative analysis capability and is difficult to identify systemic risks, the static algorithm cannot adapt to parameter drift caused by battery aging, the early identification sensitivity of gradual faults such as lithium extraction is insufficient, the black-box deep learning model lacks operable fault attribution capability, and the calculation delay of complex algorithms cannot meet the real-time requirements of the production line, which seriously restricts the test efficiency and reliability. SUMMARY

[0003] To solve the above technical problems, the full-automatic battery cell test platform and the test process optimization method thereof are provided, which solve the above problems.

[0004] To achieve the above purposes, the technical scheme adopted by the present application is as follows:

[0005] The full-automatic battery cell test platform and the test process optimization method thereof comprise:

[0006] S1, obtaining historical battery cell test big data of each power battery type, and establishing a battery cell test parameter database of each power battery type;

[0007] S2, based on the battery cell test parameter database of each power battery type, analyzing the battery cell test parameter deviation state of each power battery type, and evaluating the battery cell local fault trend of each power battery type;

[0008] S3, based on the battery cell test parameter database of each power battery type, screening the battery cell test parameters of the same power battery type for deviation state analysis, and evaluating the battery cell global fault trend of the same power battery type;

[0009] S4, determining the battery cell global fault attention factors of each power battery type according to the similarity between the battery cell local fault trend of each power battery type and the battery cell global fault trend of the same power battery type;

[0010] S5, using the battery cell global fault attention factors of each power battery type to compensate for the battery cell local fault trend of each power battery type, and generating a battery cell local fault risk probability of each power battery type.

[0011] Preferably, the standardized running parameters of the battery cells of the known power battery types are determined, and a running standard range parameter array of the battery cells of the known power battery types is established;

[0012] The normalized formula is used to normalize the cell test parameters in the cell test parameter database of each power battery type;

[0013] The sliding window segmentation method is used to divide the cell test parameter database of each power battery type according to a fixed length window per unit time, and the cell test parameter time series data of each power battery type is obtained;

[0014] Based on IsolationForest, using the known cell running standard range parameter array A = [X 11 ,X 12 ,…,X ij ] of each power battery type, wherein X ij is the known jth cell running standard range parameter of the ith power battery type, according to the known cell running standard range parameter type, a plurality of cell running parameter anomaly detection trees are pre-trained;

[0015] The cell test parameter time series data of each power battery type is substituted into the corresponding cell running parameter anomaly detection tree, and the edge tree of the cell test parameter of each power battery type from the root node to the leaf node in the tree is marked every unit time, and the path length of the cell test parameter of each power battery type is determined;

[0016] According to the path length of the cell test parameter of each power battery type, the average value of the path length of each unit time is calculated, and the average value of the path length of the cell test parameter of each power battery type is obtained;

[0017] The average value of the path length of the cell test parameter of each power battery type is substituted into the IsolationForest anomaly score function, and the abnormal score of the cell test parameter of each power battery type per unit time is obtained, as follows:

[0018]

[0019] In the formula, S ij (t) is the abnormal score of the jth cell test parameter of the ith known power battery type per unit time, T is the total length of the sliding window, h k (X ij ) is the path length of the jth cell test parameter of the ith known power battery type in the tth unit time in the kth cell running parameter anomaly detection tree, ln(·) is the natural logarithm, α is the Euler constant, and l is the total number of cell running parameter anomaly detection trees;

[0020] Based on the Euclidean distance algorithm, the space recursive cumulative distance of the test parameter time series data of each power battery type and the known standard range parameter array of each power battery type in unit time is calculated recursively to form the cumulative distance matrix of the test parameters of each power battery type.

[0021] Based on the DTW dynamic time warping algorithm, the minimum distance in the cumulative distance matrix of the test parameters of each power battery type in unit time is calculated to form the cumulative shape difference matrix of the test parameters of each power battery type.

[0022] According to the cumulative minimum distance of the test parameters of each power battery type in the cumulative shape difference matrix of the test parameters of each power battery type and the abnormal score of the test parameters of each power battery type in unit time, the feature fusion is performed by the exponentially weighted moving average to obtain the comprehensive abnormal vector of the test of each power battery type in unit time, and the method is as follows:

[0023]

[0024] In the formula, B ij (t) is the jth test comprehensive abnormal vector of the ith power battery type in the tth unit time, η is an adjustment coefficient, min(d(Q ij ,X ij )) is the cumulative minimum distance of the test parameters of each power battery type in the cumulative shape difference matrix of the test parameters of each power battery type.

[0025] The comprehensive abnormal vector of the test of each power battery type in unit time is smoothed to obtain the comprehensive abnormal trend vector of the test of each power battery type in unit time, and the method is as follows:

[0026] C ij (t) = λ·B ij (t) + (1-λ)B ij (t-1)

[0027] In the formula, C ij (t) is the jth test comprehensive abnormal trend vector of the ith power battery type in the tth unit time, λ is a decay factor, B ij (t-1) is the smoothed value of the jth test comprehensive abnormal vector of the ith power battery type in the t-1th unit time.

[0028] Preferably, according to the comprehensive abnormal trend vector of the test of each power battery type in unit time, the trend derivative of the test of each power battery type in unit time is determined.

[0029] According to the prior probability of the test parameter fluctuation of the corresponding battery cell of the known gradual failure and sudden failure of the battery cell, the gradual failure range threshold and the sudden failure range threshold of the battery cell are determined;

[0030] According to the gradual failure range threshold and the sudden failure range threshold of the battery cell, the test trend derivative of the battery cell of each power battery type in a unit time is divided by using the GMM Gaussian mixture clustering, and a test failure type trend label vector of the battery cell of each power battery type in a unit time is obtained;

[0031] Based on the test failure type trend label vector of the battery cell of each power battery type in a unit time and the comprehensive abnormal vector of the battery cell of each power battery type in a unit time, a test failure type joint vector of the battery cell of each power battery type in a unit time is established;

[0032] Based on the Logistic logistic regression, the known battery cell failure type is taken as label data, and the test failure type joint vector of the battery cell of each power battery type in a unit time is taken as feature data, the data corresponding to the known battery cell failure type of the test failure type joint vector is classified, and the local failure probability of the battery cell of each power battery type is determined.

[0033] Preferably, based on the battery cell test parameter database of each power battery type, the data is divided by using the sliding window method according to the battery cell test parameters of the same power battery type, and the time sequence data of the battery cell test parameters of the same power battery type is obtained;

[0034] According to the time sequence data of the battery cell test parameters of the same power battery type, the deviation univariate Gaussian distribution of the battery cell test parameters of the same power battery type from the known battery cell running standard range parameter array is simulated by using variable Gaussian distribution, and the deviation joint probability density of the battery cell test parameters of the same power battery type is iteratively accumulated in the following manner:

[0035]

[0036] In the formula, P(X i'j ) is the deviation joint probability density of the jth battery cell test parameter of the ith same power battery type, is a normalization factor, is a Gaussian function constant term, σ j is the standard deviation of the jth battery cell running standard range parameter of the known each power battery type, μ j is the mean value of the jth battery cell running standard range parameter of the known each power battery type, and exp() is an exponential function.

[0037] Preferably, the deviation joint probability density of the battery cell test parameters of the same power battery type is normalized;

[0038] Based on the Mahalanobis distance, the deviation of the joint probability density of the test parameters of the same power battery type is calculated, and the spatial distance between the gradual failure range threshold and the sudden failure range threshold is determined, and the deviation of the test parameters of the same power battery type is determined.

[0039] Using the analytic hierarchy process, the weight of the joint probability density of the deviation of the test parameters of the same power battery type is given in unit time.

[0040] Based on the weight of the joint probability density of the deviation of the test parameters of the same power battery type in unit time and the joint probability density of the deviation of the test parameters of the same power battery type, the total failure probability of the deviation of the test parameters of the same power battery type in unit time is calculated.

[0041] Based on the LSTM long short-term memory network, the total failure probability of the deviation of the test parameters of the same power battery type in unit time is used as a basic constant, the time series data of the test parameters of the same power battery type is used as feature data input, and the global failure probability of the same power battery type is used as output.

[0042] Preferably, based on the local failure probability of each power battery type, a local failure probability vector array of each power battery type is formed.

[0043] Based on the global failure probability of the same power battery type, a global failure trend matrix of the same power battery type is formed.

[0044] Using the DTW dynamic time warping algorithm, the shape similarity between each element in the global failure trend matrix of the same power battery type and the local failure probability vector array of each power battery type is calculated and normalized to obtain the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability in unit time, in the following manner:

[0045]

[0046] In the formula, E j is the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability, DTW(G ij ,H i'j ) is the shape similarity between the local failure probability of each power battery type in the global failure trend matrix of the same power battery type and the local failure probability vector array of each power battery type, G ij (t1) is the local failure probability of the jth battery of the ith power battery type in the t1th unit time, H i'j(t2) is the jth battery cell global failure probability of the ith same power battery type in the t2th unit time.

[0047] Preferably, according to the unit time, the cell local failure probability vector of each power battery type and the cell global failure probability normalization similarity score are marked as the cell test parameter of the same power battery type in the corresponding unit time, and the cell local failure correlation cell test global parameter of each power battery type is marked as the cell local failure correlation cell test global parameter of each power battery type.

[0048] Based on the known cell local failure probability of each power battery type, the Mahalanobis distance contribution degree of the cell local failure correlation cell test global parameter of each power battery type to the cell local failure type is verified, and the known cell local failure correlation cell test global parameter contribution weight of each power battery type is obtained.

[0049] Based on the unit time, the cell local failure probability vector of each power battery type and the cell global failure probability normalization similarity score and the known cell local failure correlation cell test global parameter contribution weight of each power battery type, the known cell local failure correlation cell test global parameter factor vector of each power battery type is calculated in the following manner:

[0050]

[0051] Wherein, F is the known cell local failure correlation jth cell test global parameter factor vector of each power battery type, W j is the known cell local failure correlation jth cell test global parameter contribution weight of each power battery type, and n is the total number of cell test global parameters.

[0052] Preferably, based on the weighted compensation formula, the known cell local failure correlation cell test global parameter factor vector of each power battery type is used to correct the cell local failure probability of each power battery type, and the known cell local-global failure compensation probability of each power battery type is obtained in the following manner:

[0053]

[0054] Wherein, is the jth cell local-global failure compensation probability of the ith known power battery type in the tth unit time, is the jth cell test local standardized parameter deviation value of the ith known power battery type in the tth unit time.

[0055] Based on the known partial-global fault compensation probability of each power battery type of the battery cell and the known partial fault of each power battery type of the battery cell, a global parameter factor vector of the battery cell test is substituted into the Logistic logistic regression, and the partial fault probability of each power battery type of the battery cell is output, in the following manner:

[0056]

[0057] Wherein, GS i is the partial fault probability of the i-th power battery type of the battery cell, β0 is the intercept, β1 and β2 are linear regression coefficients;

[0058] Determine whether the partial fault probability of each power battery type of the battery cell is in the battery tolerable range threshold interval, if yes, determine that the battery test is passed, if not, determine that the battery test is not passed.

[0059] Compared with the prior art, the beneficial effects of the present application are as follows:

[0060] The present application proposes a full-automatic battery cell test bench and its test process optimization scheme, which realizes a breakthrough improvement in power battery fault detection by constructing a multi-level intelligent analysis system; adopts a local-global collaborative analysis mechanism, significantly improves the early fault detection rate and greatly reduces the false alarm rate; introduces a dynamic weight compensation algorithm, which can intelligently identify key fault inducements such as temperature and internal resistance, greatly improves the diagnosis accuracy, is compatible with multiple battery systems, improves the battery test efficiency, and provides an intelligent solution for power battery quality management. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 It is a full-automatic battery cell test bench and its test process optimization method flow chart. DETAILED DESCRIPTION

[0062] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and other obvious variants can be conceived by those skilled in the art.

[0063] Referring to Figure 1 The full-automatic battery cell test bench and its test process optimization method, as shown in the figure, comprises:

[0064] S1, obtaining historical battery test big data of each power battery type, and establishing a battery test parameter database of each power battery type;

[0065] S2, based on the battery test parameter database of each power battery type, analyzing the deviation state of the battery test parameters of each power battery type, and evaluating the partial fault trend of each power battery type of the battery cell;

[0066] The step S2 includes the following contents:

[0067] The standard operation parameters of the battery cells of each known power battery type are determined, and a standard range parameter array of the operation of the battery cells of each known power battery type is established.

[0068] The battery cell test parameters in the battery cell test parameter database of each power battery type are normalized by using a normalization formula;

[0069] The battery cell test parameter database of each power battery type is divided according to a fixed length window of unit time by using a sliding window segmentation method, and time sequence data of the battery cell test parameters of each power battery type is obtained.

[0070] Based on IsolationForest, a plurality of battery cell operation parameter anomaly detection trees are pre-trained according to the standard range parameter array A = [X 11 ,X 12 ,…,X ij ] of the operation of the battery cells of each known power battery type, where X ij is the jth battery cell operation standard range parameter of the ith known power battery type.

[0071] The battery cell test parameter time sequence data of each power battery type is substituted into the corresponding battery cell operation parameter anomaly detection tree, and the edge tree of the battery cell test parameter of each power battery type from the root node to the leaf node in the tree is marked in each unit time, and the path length of the battery cell test parameter of each power battery type is determined.

[0072] According to the path length of the battery cell test parameter of each power battery type, the average value of the path length in each unit time is counted, and the average value of the path length of the battery cell test parameter of each power battery type is obtained.

[0073] The average value of the path length of the battery cell test parameter of each power battery type is substituted into the IsolationForest anomaly score function, and the anomaly score of the battery cell test parameter of each power battery type in each unit time is obtained, and the method is as follows:

[0074]

[0075] In the formula, S ij (t) is the anomaly score of the jth battery cell test parameter of the ith known power battery type in the tth unit time, T is the total length of the sliding window, and h k (X ij) is the path length of the jth battery cell test parameter of the ith power battery type in the tth unit time in the abnormal detection tree of the battery cell operation parameter, ln(·) is a natural logarithm, α is Euler's constant, and l is the total number of the battery cell operation parameter abnormal detection trees;

[0076] Based on the Euclidean distance algorithm, the spatial recursive cumulative distance of the battery cell test parameter time series data of each power battery type and the battery cell operation standard range parameter array of each power battery type in a unit time is recursively calculated to form a battery cell test parameter cumulative distance matrix of each power battery type;

[0077] Based on the DTW dynamic time warping algorithm, the minimum distance in the battery cell test parameter cumulative distance matrix of each power battery type in a unit time is calculated to form a battery cell test parameter cumulative shape difference matrix of each power battery type;

[0078] According to the battery cell test parameter cumulative minimum distance of each power battery type in the battery cell test parameter cumulative shape difference matrix of each power battery type and the abnormal score of the battery cell test parameter of each power battery type in a unit time, feature fusion is performed by using an exponentially weighted moving average to obtain a battery cell test comprehensive abnormal vector of each power battery type in a unit time, in the following manner:

[0079]

[0080] In the formula, B ij (t) is the jth battery cell test comprehensive abnormal vector of the ith power battery type in the tth unit time, η is an adjustment coefficient, min(d(Q ij ,X ij )) is the battery cell test parameter cumulative minimum distance of each power battery type in the battery cell test parameter cumulative shape difference matrix of each power battery type;

[0081] The battery cell test comprehensive abnormal vector of each power battery type in a unit time is smoothed to obtain a battery cell test comprehensive abnormal trend vector of each power battery type in a unit time, in the following manner:

[0082] C ij (t) = λ·B ij (t) + (1-λ)B ij (t-1)

[0083] In the formula, C ij (t) is the jth battery cell test comprehensive abnormal trend vector of the ith power battery type in the tth unit time, λ is a decay factor, B ij (t-1) is a smoothed value of the jth battery cell test comprehensive abnormal vector of the ith power battery type in the t-1th unit time.

[0084] Step S2 also includes:

[0085] According to the comprehensive abnormal trend vector of the battery cell test of each power battery type per unit time, the trend derivative of the battery cell test of each power battery type per unit time is determined;

[0086] According to the prior probability of the parameter fluctuation of the battery cell gradual failure and the corresponding battery cell test, the battery cell gradual failure range threshold and the sudden failure range threshold are determined;

[0087] According to the battery cell gradual failure range threshold and the sudden failure range threshold, the trend derivative of the battery cell test of each power battery type per unit time is divided by using the GMM Gaussian mixture clustering, to obtain the battery cell test fault type trend label vector of each power battery type per unit time;

[0088] Based on the battery cell test fault type trend label vector of each power battery type per unit time and the battery cell test comprehensive abnormal vector per unit time, the battery cell test fault type joint vector of each power battery type per unit time is formed;

[0089] Based on the Logistic logistic regression, the known battery cell failure type is taken as the label data, and the battery cell test fault type joint vector of each power battery type per unit time is taken as the feature data, the data corresponding to the known battery cell failure type is classified according to the battery cell test fault type joint vector, and the local failure probability of each power battery type is determined.

[0090] In use, the contents in the above steps are combined,

[0091] As further content, through numerical + morphological dual verification, false positives caused by single sensor false alarms are avoided; the sensitivity of DTW to curve morphology can identify early lithium precipitation (≥5 cycle periods in advance) that cannot be detected by traditional threshold methods; the combination of sliding window + Isolation Forest meets the real-time requirement (single cell detection delay <50ms), and then the template matching of DTW only compares the key feature dimensions (such as voltage + temperature), reducing the calculation load; the logistic regression feedback automatically optimizes the GMM clustering center, adapting to the parameter drift caused by battery aging; the output fault type label joint vector (fault type, confidence, recommended measures) can be directly applied to maintenance decision. The present scheme combines Isolation Forest and DTW for battery cell detection, solves the existing battery cell test data fragmentation alarm problem through EWMA trend coupling, and the two-level classification system of GMM clustering + logistic regression replaces the fixed threshold, adapting to different attenuation modes (such as sudden mutation caused by fast charging vs. gradual change caused by cycle aging).

[0092] S3, based on the battery type of each cell test parameter database, screening the same power battery type of cell test parameters for deviation state analysis, evaluation of the same power battery type of cell global failure trend;

[0093] The step S3 includes the following contents:

[0094] Based on the battery type of each cell test parameter database, according to the same power battery type of cell test parameters, the data is divided by using the sliding window method, and the same power battery type of cell test parameter time series data is obtained;

[0095] According to the same power battery type of cell test parameter time series data, the deviation of the same power battery type of cell test parameter from the known each power battery type of cell running standard range parameter array is simulated by using variable Gaussian distribution, and the deviation of the same power battery type of cell test parameter from the known each power battery type of cell running standard range parameter array is simulated by using variable Gaussian distribution. Iterative accumulation is the same power battery type of cell test parameter deviation joint probability density, as follows:

[0096]

[0097] In the formula, P(X i'j ) is the deviation of the jth cell test parameter of the ith same power battery type from the joint probability density, is a normalization factor, is a Gaussian function constant term, σ j is the standard deviation of the jth cell running standard range parameter of the known each power battery type, μ j is the mean of the jth cell running standard range parameter of the known each power battery type, exp() is the exponential function;

[0098] Step S3 also includes:

[0099] The same power battery type of cell test parameter deviation joint probability density is normalized;

[0100] Based on the Mahalanobis distance, the space distance between the same power battery type of cell test parameter deviation joint probability density normalized value and the gradual failure range threshold and the sudden failure range threshold is calculated, and the same power battery type of cell test parameter deviation failure type joint probability density is determined;

[0101] Using the analytic hierarchy process, the same power battery type of cell test parameter deviation failure type joint probability density weight is given in unit time;

[0102] Based on the joint probability density weight of the same type of power battery cell test parameter deviation fault type and the joint probability density of the same type of power battery cell test parameter deviation in unit time, the total fault probability of the same type of power battery cell test parameter deviation fault type in unit time is calculated.

[0103] Based on the LSTM long short-term memory network, the total fault probability of the same type of power battery cell test parameter deviation fault type in unit time is used as a basic constant, the time series data of the same type of power battery cell test parameter is used as feature data input, and the global fault probability of the same type of power battery cell is used as output.

[0104] In use, the contents in the above steps are combined,

[0105] As further content, the time series data (voltage, temperature, etc.) of the same type of cell is divided into a fixed time window (such as 60 seconds), local features (mean, variance, etc.) are extracted, the data continuity problem is solved, the joint probability density of each window data is calculated, and is scaled to the [0, 1] interval to eliminate the dimension difference of the parameters, facilitate cross-battery type comparison, and then use Mahalanobis distance to verify the spatial distance of the joint probability density of each window data to the fault threshold, determine the joint probability density of the same type of power battery cell test parameter deviation fault type, and use the analytic hierarchy process to give the joint probability density of the same type of power battery cell test parameter deviation fault type weight (such as the contribution of temperature to thermal runaway > voltage fluctuation), reflecting the sensitivity of different parameters to known fault types; by calculating the total fault probability of the same type of power battery cell test parameter deviation fault type in unit time as a single constant, the attention mechanism significantly reduces the calculation complexity for reverse positioning of high-contribution parameters;

[0106] S4, according to the similarity between the local fault trend of each type of power battery cell and the global fault trend of the same type of power battery cell, determine the global fault attention factor of each type of power battery cell;

[0107] The step S4 includes the following contents:

[0108] Based on the local fault probability of each type of power battery cell, a local fault probability vector array of each type of power battery cell is formed;

[0109] Based on the global fault probability of the same type of power battery cell, a global fault trend matrix of the same type of power battery cell is formed;

[0110] The DTW dynamic time warping algorithm is used to calculate the shape similarity between each element in the global failure trend matrix of the same power battery type and the local failure probability vector array of each power battery type, and to perform normalization processing, to obtain the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability of the battery under unit time, in the following manner:

[0111]

[0112] In the formula, E j is the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability of the battery, DTW(G ij , H i'j ) is the shape similarity between the local failure probability of each power battery type in the global failure trend matrix of the same power battery type and the local failure probability vector array of each power battery type, G ij (t1) is the local failure probability of the jth battery of the ith power battery type under the t1th unit time, H i'j (t2) is the global failure probability of the jth battery of the ith' same power battery type under the t2th unit time.

[0113] As further content, to avoid misunderstanding by those skilled in the art, t1 and t2 are essentially the step length of the unit time window, but in order to distinguish the difference between local and global, t1 is used to represent the local unit time window step length, and t2 is used to represent the global unit time window step length. The length corresponding to the code should be the step length of each loop, which is well known to those skilled in the art, and will not be explained here.

[0114] Step S4 further comprises:

[0115] According to the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability of the battery under unit time, the corresponding unit time is marked for the same power battery type battery test parameter, and is recorded as the local failure correlation battery test global parameter of each power battery type.

[0116] Based on the known local failure probability of each power battery type, the Mahalanobis distance contribution degree of the local failure correlation battery test global parameter of each power battery type to the local failure type is normalized to obtain the contribution weight of the known local failure correlation battery test global parameter of each power battery type.

[0117] Based on the unit time of each power battery type of cell local fault probability vector and cell global fault probability normalized similarity score and known each power battery type of cell local fault associated cell test global parameter contribution weight, the known each power battery type of cell local fault associated cell test global parameter factor vector is calculated, in the following way:

[0118]

[0119] Wherein, F is the known each power battery type of cell local fault associated jth cell test global parameter factor vector, W j The known each power battery type of cell local fault associated jth cell test global parameter contribution weight, n is the total number of cell test global parameters;

[0120] In use, combined with the content in the above steps,

[0121] As further content, establish local fault probability vector array and global fault trend matrix, ensure the quantification of single cell in different time window abnormal degree, retain the time sequence dependence, reflect the common fault mode of group cells, eliminate individual random fluctuation, use morphological similarity analysis to solve the time axis shift problem caused by different cells due to aging speed difference (such as the fault peak of cell A is delayed than cell B), extract the cell test parameters of similarity, and improve the possibility of parameter abnormality of high similarity group reflecting systematic risk (such as some batch electrolyte pollution) coarse granularity; And fault parameter contribution degree verification, identify the parameters with the greatest impact on the overall deviation (such as temperature contribution degree 50% → cooling system failure) combined with similarity and contribution degree, generate global attention factor vector, and highlight the common fault inducement. This scheme realizes the leap from single cell anomaly detection to global systematic risk warning through three-stage analysis of time sequence alignment-group mining-parameter tracing.

[0122] S5, using the global fault attention factor of each power battery type of cell, compensating for the local fault trend of each power battery type of cell, generating the local fault risk probability of each power battery type of cell;

[0123] The step S5 includes the following content:

[0124] Based on the weighted compensation formula, the known each power battery type of cell local fault associated cell test global parameter factor vector is used to correct the local fault probability of each power battery type of cell, and the known each power battery type of cell local-global fault compensation probability is obtained, in the following way:

[0125]

[0126] Wherein, is the local-global fault compensation probability of the jth cell of the ith power battery type in the tth unit time, is the test local normalized parameter deviation value of the jth cell of the ith power battery type in the tth unit time;

[0127] Based on the known cell local-global fault compensation probability of each power battery type and the known cell local fault associated cell test global parameter factor vector of each power battery type, the Logistic logistic regression is substituted, and the cell local fault probability of each power battery type is output, in the following manner:

[0128]

[0129] Wherein, GS i is the cell local fault probability of the ith power battery type, β0 is the intercept, β1 and β2 are linear regression coefficients;

[0130] Determine whether the cell local fault probability of each power battery type is in the battery tolerable range threshold interval, if yes, determine that the battery test is passed, if not, determine that the battery test is not passed.

[0131] In use, the contents in the above steps are combined,

[0132] As further content, through global-local collaborative analysis and dynamic risk compensation mechanism, the precision and intelligent upgrading of power battery fault detection are realized. The core principle is: based on the extracted global fault attention factors (such as the weight distribution of parameters such as temperature and internal resistance), the generated local fault probability is weighted and compensated, and then the Logistic regression model is fused with multi-dimensional features, and finally the cell health state is determined in combination with the dynamic threshold. This process significantly improves the comprehensive performance of the detection system through "group knowledge correction of individual deviation" and "multi-parameter joint decision".

[0133] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A full-automatic battery cell test platform and a test process optimization method thereof, characterized in that, The method comprises the following steps: S1, obtaining historical big data of battery cell test of each power battery type, and establishing a battery cell test parameter database of each power battery type; S2, based on the battery cell test parameter database of each power battery type, analyzing the deviation state of the battery cell test parameters of each power battery type, and evaluating the local fault trend of the battery cell of each power battery type; S3, based on the battery cell test parameter database of each power battery type, screening the battery cell test parameters of the same power battery type for deviation state analysis, and evaluating the global fault trend of the battery cell of the same power battery type; S4, determining the global fault attention factors of the battery cell of each power battery type according to the similarity between the local fault trend of the battery cell of each power battery type and the global fault trend of the battery cell of the same power battery type; S5, using the global fault attention factors of the battery cell of each power battery type to compensate for the local fault trend of the battery cell of each power battery type, and generating the local fault risk probability of the battery cell of each power battery type.

2. The fully automatic cell testing platform and its testing procedure optimization method according to claim 1, characterized in that, The step S2 comprises the following contents: determining the known battery cell standardized operation parameters of each power battery type, and establishing the known battery cell operation standard range parameter array of each power battery type; using a normalization formula to normalize the battery cell test parameters in the battery cell test parameter database of each power battery type; using a sliding window segmentation method, dividing the battery cell test parameter database of each power battery type according to a fixed length window in unit time, and obtaining the time sequence data of the battery cell test parameters of each power battery type; Based on IsolationForest, using the known standard range parameter array A = [X 11 ,X 12 ,…,X ij ] of each power battery type, wherein X ij is the jth battery cell operating standard range parameter of the ith power battery type, pre-training several battery cell operating parameter anomaly detection trees according to the known battery cell operating standard range parameter type of each power battery type; substituting the time sequence data of the battery cell test parameters of each power battery type into the corresponding battery cell operation parameter anomaly detection tree, marking the edge tree of the battery cell test parameters of each power battery type from the root node to the leaf node in the tree in each unit time, and determining the path length of the battery cell test parameters of each power battery type; according to the path length of the battery cell test parameters of each power battery type, calculating the average value of the path length in each unit time, and obtaining the average value of the path length of the battery cell test parameters of each power battery type; substituting the average value of the path length of the battery cell test parameters of each power battery type into the isolation forest anomaly score function, obtaining the anomaly score of the battery cell test parameters of each power battery type in unit time, and the method is as follows: In the formula, S ij (t) is the abnormal score of the jth cell test parameter of the ith power battery type in the tth unit time, T is the total length of the sliding window, h k (X ij ) is the path length of the jth cell test parameter of the ith power battery type in the tth unit time in the kth cell operating parameter anomaly detection tree, ln(·) is a natural logarithm, and a is Euler's constant. l is the total number of cell operating parameter anomaly detection trees. based on the Euclidean distance algorithm, recursively calculating the spatial recursive cumulative distance of the battery cell test parameter time sequence data of each power battery type and the known battery cell operation standard range parameter array of each power battery type in unit time, and establishing the battery cell test parameter cumulative distance matrix of each power battery type; based on the DTW dynamic time warping algorithm, calculating the minimum distance in unit time in the battery cell test parameter cumulative distance matrix of each power battery type, and establishing the battery cell test parameter cumulative shape difference matrix of each power battery type; The feature fusion is performed on the cumulative minimum distance of the battery cell test parameters of each battery type in the cumulative form difference matrix of the battery cell test parameters of each battery type and the abnormal score of the battery cell test parameters of each battery type per unit time according to the exponentially weighted moving average, to obtain a battery cell test comprehensive abnormal vector of each battery type per unit time, in the following manner: In the formula, B ij (t) is the jth battery cell test comprehensive abnormality vector of the ith power battery type in the tth unit time, η is an adjustment coefficient, min(d(Q ij ,X ij )) is the cumulative minimum distance of the battery cell test parameters of each power battery type in the cumulative form difference matrix of the battery cell test parameters of each power battery type. The battery cell test comprehensive abnormal vector of each battery type per unit time is smoothed to obtain a battery cell test comprehensive abnormal trend vector of each battery type per unit time, in the following manner: C ij (t) = λ · B ij (t) + (1 - λ) B ij (t - 1) In the formula, C ij (t) is the jth battery cell test comprehensive abnormal trend vector of the ith power battery type in the tth unit time, λ is a decay factor, B ij (t-1) is the jth battery cell test comprehensive abnormal vector smooth value of the ith power battery type in the t-1th unit time.

3. The fully automatic electric cell test bench and its test procedure optimization method according to claim 2, characterized in that, The step S2 further includes: A battery cell test trend derivative of each battery type per unit time is determined according to the battery cell test comprehensive abnormal trend vector of each battery type per unit time; A gradual fault range threshold and a sudden fault range threshold of the battery cell are determined according to the prior probability of the fluctuation of the battery cell test parameters corresponding to the gradual fault and the sudden fault of the known battery cell; The battery cell test trend derivative of each battery type per unit time is divided according to the gradual fault range threshold and the sudden fault range threshold by using GMM Gaussian mixture clustering, to obtain a battery cell test fault type trend label vector of each battery type per unit time; A battery cell test fault type joint vector of each battery type per unit time is formed based on the battery cell test fault type trend label vector of each battery type per unit time and the battery cell test comprehensive abnormal vector of each battery type per unit time; The data of the known battery cell fault type corresponding to the battery cell test fault type joint vector is classified based on Logistic logistic regression, with the known battery cell fault type as label data and the battery cell test fault type joint vector of each battery type per unit time as feature data, to determine the battery cell local fault probability of each battery type.

4. The fully automatic electric cell test bench and its test procedure optimization method according to claim 3, characterized in that, The step S3 includes the following contents: The battery cell test parameter time series data of the same battery type is obtained by dividing the data according to the battery cell test parameters of the same battery type by using the sliding window method based on the battery cell test parameter database of each battery type. The deviation univariate Gaussian distribution of the battery cell test parameters of the same battery type from the known battery cell operating standard range parameter array is simulated according to the battery cell test parameter time series data of the same battery type by using the variable Gaussian distribution, which is iteratively accumulated as the battery cell test parameter deviation joint probability density of the same battery type, in the following manner: where P(X i'j ) is the joint probability density of the jth cell test parameter deviation of the ith homogenous battery type, is a normalization factor, is a Gaussian function constant term, σj is the standard deviation of the jth cell operating standard range parameter of each known homogenous battery type, μ j is the mean of the jth cell operating standard range parameter of each known homogenous battery type, and exp() is the exponential function.

5. The fully automatic electric cell test bed and its test procedure optimization method according to claim 4, characterized in that, The step S3 further includes: The battery cell test parameter deviation joint probability density of the same battery type is normalized; The battery cell test parameter deviation fault type joint probability density of the same battery type is determined based on the Mahalanobis distance, which calculates the spatial distance between the normalized value of the battery cell test parameter deviation joint probability density of the same battery type and the gradual fault range threshold and the sudden fault range threshold; The weight of the battery cell test parameter deviation fault type joint probability density of the same battery type per unit time is given by using the analytic hierarchy process. Based on the joint probability density weight of the same power battery type cell test parameter deviation fault type and the same power battery type cell test parameter deviation joint probability density in unit time, the total fault probability of the same power battery type cell test parameter deviation fault type in unit time is calculated; Based on the LSTM long short-term memory network, the total fault probability of the same power battery type cell test parameter deviation fault type in unit time is taken as the basic constant, the same power battery type cell test parameter time series data is taken as the feature data input, and the same power battery type cell global fault probability is taken as the output.

6. The fully automatic electric cell test bench and its test procedure optimization method according to claim 5, characterized in that, The step S4 includes the following contents: Based on the local fault probability of each power battery type cell, a local fault probability vector array of each power battery type cell is established; Based on the global fault probability of the same power battery type cell, a global fault trend matrix of the same power battery type cell is established; Using the DTW dynamic time warping algorithm, the shape similarity between each element in the global fault trend matrix of the same power battery type cell and the local fault probability vector array of each power battery type cell is calculated and normalized to obtain the normalized similarity score of the local fault probability vector of each power battery type cell and the global fault probability in unit time, in the following manner: wherein E j is the normalized similarity score of the local failure probability vector of each power battery type and the global failure probability of the cell, DTW(G ij , H i'j ) is the shape similarity between the local failure probability of each power battery type in the global failure trend matrix of the cell of the same power battery type and the local failure probability vector array of each power battery type, G ij (t1) is the local failure probability of the jth cell of the ith power battery type at the t1th unit time, H i'j (t2) is the global failure probability of the jth cell of the ith' same power battery type at the t2th unit time.

7. The fully automatic electric cell test bench and its test procedure optimization method according to claim 6, characterized in that, The step S4 further includes: According to the normalized similarity score of the local fault probability vector of each power battery type cell and the global fault probability in unit time, the same power battery type cell test parameter in the corresponding unit time is marked as the local fault associated global parameter of each power battery type cell test; Based on the known local fault probability of each power battery type cell, the Mahalanobis distance contribution degree of the local fault associated global parameter of each power battery type cell test to the local fault type of the cell is normalized to obtain the contribution weight of the known local fault associated global parameter of each power battery type cell test; Based on the normalized similarity score of the local fault probability vector of each power battery type cell and the global fault probability in unit time and the contribution weight of the known local fault associated global parameter of each power battery type cell test, a known local fault associated global parameter factor vector of each power battery type cell test is calculated in the following manner: Wherein, F is the partial failure of the battery cell associated with the jth battery cell test global parameter factor vector of each known battery type, W j is the partial failure of the battery cell associated with the jth battery cell test global parameter contribution weight of each known battery type, and n is the total number of battery cell test global parameters.

8. The fully automatic electric cell test bed and its test procedure optimization method according to claim 7, characterized in that, The step S5 includes the following contents: Based on the weighted compensation formula, the known local fault associated global parameter factor vector of each power battery type cell test is used to correct the local fault probability of each power battery type cell to obtain the known local-global fault compensation probability of each power battery type cell, in the following manner: wherein, is the known partial-global fault compensation probability of the jth cell of the ith power battery type at the tth unit time, is the known test partial normalized parameter deviation of the jth cell of the ith power battery type at the tth unit time; Based on the known local-global fault compensation probability of each power battery type cell and the known local fault associated global parameter factor vector of each power battery type cell, the Logistic logistic regression is substituted to output the local fault probability of each power battery type cell, in the following manner: wherein GS i is the local fault probability of the i-th power battery type, β0 is the intercept, and β1 and β2 are linear regression coefficients. Judge the local fault probability of each power battery type, whether it is in the battery tolerable range threshold interval, if yes, judge that the battery test is passed, if not, determine that the battery test is failed.

Images