阅读说明：本技术 退役锂离子电池和电池模块的健康状态评估 (State of health assessment of decommissioned lithium ion batteries and battery modules ) 是由 许敏洁 孙耀峰 于 2021-06-21 设计创作，主要内容包括：旧电池从第一开路电压(OCV1)到第二OCV2进行短时间放电,并测量放电电流△Q。OCV1被输入到校准曲线模型中,以获得第一建模荷电状态(SOC1)值,OCV2被输入到校准曲线模型中以获得第二建模SOC2值。健康状态(SOH)的计算公式为△Q/[Q-(new) x(SOC1–SOC2)],其中Q-(new)是新电池容量。根据SOH值对旧电池进行分类,以进行再利用或处置。校准曲线模型是通过人工智能(AI)对旧电池完全充电和放电的OCV、SOC数据点进行建模获得的。仅对一个目标区域内的具有较低的SOC一阶导数的OCV值进行建模,并且OCV1和OCV2在该目标区域内。(The old battery was discharged for a short time from the first open voltage (OCV1) to the second OCV2, and the discharge current Δ Q was measured. The OCV1 is input into a calibration curve model to obtain a first modeled state-of-charge (SOC1) value, and the OCV2 is input into a calibration curve model to obtain a second modeled SOC2 value. The calculation formula of the state of health (SOH) is delta Q/[ Q [ new x(SOC1–SOC2)]Wherein Q is new Is the new battery capacity. The used batteries are classified according to SOH values for reuse or disposal. The calibration curve model is obtained by modeling OCV, SOC data points of the old battery fully charged and discharged through Artificial Intelligence (AI). For only one target region having a lower first derivative of SOCThe OCV values were modeled and OCV1 and OCV2 were within this target region.)

1. A method of screening batteries for reuse or disposal, comprising:

measuring an Open Circuit Voltage (OCV) of the battery, i.e., OCV 1;

discharging the battery and re-measuring the OCV of the battery as OCV1 when the OCV1 is above a regional OCV limit until OCV1 is within the regional OCV limit;

discharging the battery using a constant current for a first period of time;

multiplying the value of said constant current by the value of said first time period to produce a value of Δ Q, which is stored in a computer memory;

measuring a second OCV, OCV2, of the battery after the battery has been discharged for a first period of time, after a rest period of time;

inputting OCV1 to a calibration function processor that outputs a first modeled state of charge value (SOC1) corresponding to OCV1 input to the calibration function processor;

inputting OCV2 to the calibration function processor, which outputs a second modeled state of charge value (SOC2) corresponding to OCV2 input to the calibration function processor;

generating a state of health (SOH) value by dividing the Δ Q value by a divisor, the divisor being a charge capacity of the new battery multiplied by a difference between SOC1 and SOC 2;

when the SOH value is higher than an SOH threshold, the battery is subjected to reuse classification by using the SOH value, and when the SOH value is lower than the SOH threshold, the battery is subjected to disposal classification.

2. The method of claim 1, further comprising:

the battery is pre-screened by measuring an initial voltage of the battery, and the battery is discarded when the initial voltage of the battery is below a minimum pre-screening voltage.

3. The method of claim 1, wherein the constant current does not exceed 10% of a maximum battery current.

4. The method of claim 1, further comprising:

sorting the batteries into one of a plurality of bins according to the SOH values, each bin of the plurality of bins receiving batteries having a different range of SOH values.

5. The method of claim 1, further comprising:

generating an OCV-SOC model that programs the calibration function processor to generate a modeled SOC value from the input OCV value by:

discharging an old calibration cell to a minimum voltage;

wherein the old calibration battery is of the same type as the battery, wherein the old calibration battery has an SOH value of at least 80%;

charging the old calibration battery at a constant current and recording first data points while charging the old calibration battery until a maximum voltage is reached, wherein each first data point is a (V)_chargeQ) value, wherein V_chargeIs the measured voltage of the old calibration cell, Q is the voltage from the minimum voltage to V_chargeAn integrated charge applied to the old calibration cell;

discharging the old calibration cell at a constant current and recording a second data point while the old calibration cell is discharging until the minimum voltage is reachedWherein each second data point is a (V)_dischargeQ) value, wherein V_dischargeIs the measured voltage of the old calibration cell, Q is the maximum voltage to V of the old calibration cell_dischargeThe integrated charge provided;

recording the total integrated charge of the old calibration battery discharged from the maximum voltage to the minimum voltage as a current charge capacity value Q_now；

For each pair of first and second data points having the same Q value, pair V_chargeAnd V_dischargeAveraging to obtain an average OCV for one data point;

for each data point, divide Q by Q_nowTo obtain a state of charge (SOC) value for the data point;

wherein the SOC value from the data point is a model input SOC value;

wherein the average OCV value from the data points is the model input OCV value;

storing the model input SOC value and the model input OCV value in a computer memory;

generating parameters describing the OCV-SOC model using the model input OCV value and a model input SOC value, the OCV-SOC model programming the calibration function processor to generate the modeled SOC value from the input OCV value.

6. The method of claim 5, further comprising:

generating a first derivative of the SOC value as a function of the OCV value from the data points;

comparing the first derivative of the SOC value with a threshold value, and discarding data points having a first derivative of the SOC value above the threshold value;

determining a target range of OCV values for which the first derivative of the SOC value is below the threshold;

setting the regional OCV limits at boundaries of the target range of OCV values;

wherein both OCV1 and OCV2 are within the OCV value target ranges.

7. The method of claim 5, wherein generating parameters describing the OCV-SOH model using the model input OCV value and the model input SOH value further comprises:

(m) inputting said model input OCV values to an input of a neural network;

processing said model input OCV values using said neural network to generate a calculated SOC value;

comparing said calculated SOC value to said model input SOC value using a loss function to produce a loss value;

adjusting weights of nodes within the neural network using the loss values and repeating from step (m) until a modeling endpoint is reached;

storing said weights in a computer memory coupled to said neural network;

when the modeling endpoint is reached, the modeled SOC value is generated from the OCV input using a final weight value of the neural network to implement the calibration function processor to generate the modeled SOC value from the input OCV value.

8. The method of claim 7, wherein the neural network comprises a first level node that performs a wavelet function, and a second level node that performs a product function, and a third level node that performs a summation function.

9. A battery screening method, comprising:

comparing an initial Open Circuit Voltage (OCV) of the battery to an OCV voltage target range;

discharging the battery until the OCV of the battery is within the target range when the initial OCV is greater than the OCV voltage target range;

when the initial OCV is lower than the OCV voltage target range, charging the battery until the OCV of the battery is within the target range;

the battery takes a rest after charging or discharging;

recording a first OCV of the battery, wherein the first OCV is within the target range;

discharging the battery and measuring a Δ Q charge provided by the battery when discharging from the first OCV to a second OCV;

measuring the second OCV after the battery has been discharged from the first OCV to the second OCV at rest;

inputting said first OCV to a calibration function generator that returns a first modeled state of charge (SOC) value corresponding to a first input OCV value;

inputting said second OCV to a calibration function generator that returns a second modeled state of charge (SOC) value corresponding to a second input OCV value;

calculating a modeled state of health (SOH) value using the first modeled SOC value, the second modeled SOC value, and the measured aq value;

using the modeled SOH value to determine when to discard the battery, and when to reuse the battery,

the battery is thereby screened according to the modeled SOH value as a function of the first and second OCV measurements.

10. The battery screening method of claim 9, wherein the first OCV differs from the second OCV by no more than 0.1 volts.

11. The battery screening method according to claim 9, wherein the OCV voltage target range includes OCV voltages lower than OCV voltages outside the target range,

where a lower OCV voltage was tested.

12. The battery screening method according to claim 9, further comprising:

generating a first derivative of the SOC value as a function of the OCV value from the data points;

comparing the first derivative of the SOC value with a threshold value, and discarding data points having a first derivative of the SOC value above the threshold value;

determining a target range of OCV values for which the first derivative of the SOC value is below the threshold;

a regional OCV limit is set at the boundary of the OCV value target range.

13. The battery screening method according to claim 12, further comprising:

generating calibration data points by measuring a plurality of used batteries, each of the used batteries processed by a calibration data collection process comprising:

first discharging said old battery until a lower voltage target is reached;

cooling the used battery for a period of time after the initial discharge;

charging the used battery with a constant current until a higher voltage target is reached, and recording data points during charging of the used battery, wherein each data point has a charging voltage V measured on the used battery_chargeAnd at reach V_chargeAn accumulated charge Q for charging the old battery;

discharging the used battery using the constant current until the lower voltage target is reached, and recording data points as the used battery is discharged, wherein each data point has a discharge voltage V measured on the used battery_dischargeAnd at reach V_dischargeThe accumulated charge Q remaining in the old battery;

recording a total charge Q provided by the used battery as it is discharged from the higher voltage target to the lower voltage target_now；

To V_chargeAnd V_dischargeAveraging to obtain OCVs for data points having the same accumulated charge Q;

dividing the accumulated charge Q by Q_nowObtaining SOC of the data points to form SOC and OCV data points;

each SOC, OCV data point is stored in computer memory as a calibration data point.

14. The battery screening method according to claim 13, further comprising:

a plurality of calibration data points are used as inputs to a model generator that programs the calibration function generator.

15. The battery screening method of claim 14, further comprising:

(a) inputting OCV values from a plurality of calibration data points as inputs to an input layer of a neural network, the neural network generating a plurality of calculated outputs as a function of the inputs and a plurality of weights;

(b) comparing the plurality of calculated outputs to SOC values from a plurality of calibration data points using a loss function to adjust the plurality of weights;

repeating steps (a) and (b) using the adjusted values of the plurality of weights until the loss function reaches an endpoint;

when the endpoint is reached, applying the plurality of weights to the neural network to generate a calculated output of an OCV input, the calculated output being the modeled SOC value of the calibration function generator.

16. A battery state of health (SOH) estimation method, comprising:

measuring an initial cell voltage of a cell;

comparing the initial battery voltage to a range limit voltage;

discharging the battery when the initial battery voltage is greater than the range limit voltage;

measuring an Open Circuit Voltage (OCV) of the battery, OCV 1;

discharging the battery with a constant current for a test time period, and recording the charge provided by the battery for the test time period as Δ Q;

measuring an Open Circuit Voltage (OCV), OCV2, of the battery after rest after the test period;

inputting the OCV1 to a processor to generate a first modeled state of charge value SOC 1;

inputting OCV2 to the processor to generate a second modeled state of charge value SOC 2;

multiplying the charge capacity of a new battery by the difference of SOC1 and SOC2 to produce a divisor, and then dividing Δ Q by the divisor to produce a modeled state of health (SOH);

comparing the modeled SOH to an SOH threshold as a basis for disposal or reuse of the battery classification;

accordingly, the battery is classified according to the modeled SOH determined by the two OCV voltages measured for the battery.

17. The state of health (SOH) estimation method according to claim 16, further comprising:

collecting data points from a calibration cell by measuring a charging voltage value and an SOC value during charging, and a discharging voltage value and an SOC value during discharging;

averaging the charging voltage values and the discharging voltage values of data points having the same SOC value to obtain an average OCV value;

generating a calibration curve that best fits the data points;

the processor generates the modeled SOC from the OCV input to the processor using the calibration curve.

18. The state of health (SOH) estimation method according to claim 17, further comprising: the data points are input to a neural network to generate the calibration curve.

19. The state of health (SOH) estimation method according to claim 17, further comprising:

generating a first derivative of SOC as a function of OCV, wherein SOC values and OCV values are obtained from the data points;

comparing the first derivative of the SOC to an error threshold;

the range limit voltage is selected such that OCV values below the range limit voltage have a first derivative of SOC less than the error threshold, while at least some OCV values above the range limit voltage have a first derivative above the error threshold.

20. The state of health (SOH) estimation method according to claim 17, further comprising:

initially discharging said battery at a high current at least ten times said constant current such that said initial battery voltage is below said range limit voltage.

Technical Field

The invention relates to a battery screening method, in particular to a method for screening aged or retired batteries for reuse.

Background

Batteries are widely used to power various systems. Traditionally, many battery-powered systems have lower power consumption, but recently the demand for batteries used in electric vehicles (EV's) is growing. Each electric vehicle requires a large battery pack to provide the large amount of power required to propel the electric vehicle.

Electric vehicles typically use relatively expensive lithium ion batteries. The chemicals used in such advanced batteries can present disposal problems. Toxic chemicals can leak from the waste batteries and contaminate the water supply. With the popularization of electric vehicles, with the decommissioning of EV batteries, additional burden is brought to landfill sites.

Recycling lithium ion batteries and other batteries may require the use of acids or blast furnaces, which can cause other environmental problems. The low profit margin results in unattractive battery recycling.

In particular, the EV battery pack may be replaced in advance. The recommendations of electric vehicle manufacturers may require a service shop to replace battery packs having a relatively high discharge capacity that is lower than that required to ensure EV motion performance. An EV battery pack may be removed before all of the cells are depleted. Especially for large battery packs, there may be many battery cells or battery packs with a considerable remaining service life. These battery cells may be used to power other systems that are less power demanding, such as communications and computer backup systems. Rather than discarding the replaced EV battery in a landfill or melting it, reusing the EV battery extends its useful life by 5 to 7 years, thereby providing a more sustainable and environmentally friendly method.

The availability of the waste battery may be defined by its State-of-Health (SOH) ratio. SOH refers to the current storage capacity (Q) of the battery_now) With the initial or nominal storage capacity (Q) of the battery_new) The ratio of. The storage capacity is about the discharge capacity of the battery.

In contrast, the State of charge (SOC) of a battery is the current charge (Q) of the battery_current) And a battery asFront storage capacity (Q)_now) The ratio of (a) to (b). The SOC is typically displayed to the user in% of battery charge.

Q_currentCan be measured by coulomb counting or integrating the discharge current when testing old batteries. Also can measure Q of a new battery_nowOr may use the manufacturer's specifications.

When the SOC is known, the state of health, SOH, of the battery may be represented as Q_currentAnd Q_newFunction of (c):

fig. 1 shows a prior art battery capacity test. Many variations are possible and fig. 1 is for illustration only and does not necessarily represent any particular battery test.

It may take a lot of time to accurately measure the full storage capacity of the battery. Rapid charging or discharging can heat the battery, affecting the measurement. The battery may initially have a residual charge that needs to be discharged before a capacity measurement can be made.

In step 202, the battery under test is first charged to 3.8 volts by applying a Constant Current (CC) of 1C amps, and then once a target voltage of 3.8 volts is reached, the Current is reduced to maintain a Constant Voltage (CV) or 3.8 volts. The current will drop during the CV phase until a low current value is reached, such as 0.01C, or until a period of time has elapsed.

The cell was allowed to cool for one hour before proceeding to the next step. Also, the battery may be allowed to cool for 10 minutes prior to the initial charge at step 202.

At step 204, after a cooling time of 1 hour, the battery is discharged using a Constant Current (CC) with a fixed current value of 1C. Once the voltage of the battery dropped from 3.8 volts to 2.8 volts, the discharge was stopped, the battery was allowed to rest and cool for one hour.

Then, in step 206, the battery is charged by applying a Constant Current (CC) of 1C to bring the battery voltage to a higher 4.25 volts, and when the battery voltage reaches 4.25 volts, a Constant Voltage (CV) charge is performed, the current being reduced to keep the battery voltage constant at 4.25 volts. After the charging current drops below the lower threshold, the charging is ended and the battery is allowed to rest and cool for another hour.

Finally, at step 208, the cell is slowly discharged using a Constant Current (CC) of only 5% or 0.05C of the previous discharge current. This discharge current continues until the battery voltage reaches 2.8 volts. The discharge capacity of the cell was measured by integrating the 0.05C discharge current over the time required to reach the 2.8 volt end point. This integrated current can be compared to a specified charge on a fresh battery for similar tests to calculate the SOH ratio.

When the 1C discharge of step 204 exceeds one hour, a small current of 0.05C in the discharge step 208 may take a longer time, for example, 20 hours. The total test time may exceed 26 hours, including a rest time of up to one hour in steps 202, 204, 206. This lengthy testing time is both expensive and undesirable.

Fig. 2A-2B are graphs of Open Current Voltage (OCV) of a used battery as a function of its state of charge (SOC). In fig. 2A, the OCV of an unloaded battery is relatively linear in region 502, approximately between 10% and 90% SOC. Below 10% charge and above 90% charge, the curve is not linear. Fig. 2A shows the curve of a Nickel Manganese Cobalt (NMC) cell.

Fig. 2B shows a charging curve of a lithium iron phosphate (LFP) battery. The OCV is both linear and flat in region 504, i.e., between about 10% and 90% of the charge (SOC). The different chemistries of LFP and NMC cells result in different slopes in the regions 502, 504, but still provide a linear region in the 10-90% SOC range.

Fig. 3 is a graph showing charge and discharge curves. While the battery is charging, a measuring voltage V is applied to the battery_chargeSlightly above OCV. Likewise, the measured discharge voltage V_dischargeSlightly below OCV. V_chargeAnd V_dischargeThe difference between them may be caused by the passage of current through the internal resistance of the cell. For low currents, the OCV may be approximated as V according to an internal resistance (Rint) model_chargeAnd V_dischargeAverage value of (d):

for the Rint model to work, the discharge current should be less than 5% of full current, or 0.05C. The use of such a small current requires a long discharge time when the battery is fully discharged from a fully charged state. The discharge time can be as long as 19 hours to several days. Because the battery requires time to reach chemical equilibrium and cool, accurately measuring OCV can require a long rest, even overnight. To accurately measure OCV, additional electronics may need to be added to the battery tester. Therefore, the OCV test is used to screen the old batteries for disadvantages of time and cost.

Existing rapid screening methods, such as coulometry and internal resistance methods, may suffer from such long test times. The degree of fit of the internal resistance method may be low. These methods may require complex setup.

It is desirable to have a method of screening for old batteries. It is desirable to measure the discharge capacity of the old battery over a small voltage range to speed up the test. It is desirable to be able to more quickly determine the state of health of a battery using only 2 voltage measurements, and a coulomb count between 2 voltages. It is desirable to use a pre-calibration method of Artificial Intelligence (AI) to screen old batteries more quickly.

Drawings

Fig. 1 shows a prior art battery capacity test.

Fig. 2A-2B are graphs of open circuit current voltage (OCV) versus state of charge (SOC) for used batteries.

Fig. 3 is a graph showing charge and discharge curves.

Fig. 4 shows a battery modeling curve.

Fig. 5 shows the first derivative of the cell modeling curve.

Fig. 6 shows the region of the derivative of the battery modeling curve.

FIG. 7 shows a neural network used to model a calibration curve for SOC as a function of OCV.

Fig. 8 shows training of the neural network using the measured SOC as a target to generate an old battery calibration model.

Fig. 9 shows a process of testing a used battery to obtain OCV and SOC values to model a calibration curve.

Fig. 10A-10B are methods of testing and sorting used batteries based on two (Q, OCV) data points measured during a short-zone OCV test.

Fig. 11 shows the regional OCV test in more detail.

Detailed Description

The present invention relates to improvements in battery screening. The following description is presented to enable one of ordinary skill in the art to make and use the invention in the context of a particular application and its requirements. Various modifications to the preferred embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments. Thus, the present invention is not intended to be limited to the particular embodiments shown and described, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Fig. 4 shows a battery modeling curve. The inventors have taken the OCV-SOC curve of fig. 3 and flipped the x-axis and y-axis to make SOC a function of OCV.

Fig. 5 shows the derivative of the battery modeling curve. The inventors next used the first derivative of the SOC-OCV battery modeling curve of fig. 4 (as shown by the dashed line in fig. 5). The first derivative d (SOC) shows a spike at the curve of the SOC-OCV curve.

Fig. 6 shows the region of the derivative of the battery modeling curve. The inventors have noted that the derivative may be divided into three regions. The middle region 2 has the largest derivative spike. The derivative value of region 3 gradually decreases but is still higher than region 1. The derivative value of region 1 is the smallest. Zone 1 occurs at the minimum OCV value, i.e., when the battery charge is below 10% or the SOC < 10%.

The inventors have realized that regions with lower derivative values vary less and are more amenable to modeling. Surprisingly, the smallest derivative is region 1, from 0% to 10% SOC, since the prior art OCV tests show that the flat regions 502, 504 (FIGS. 2A-2B) are the middle regions, i.e., 10-90% SOC. However, when SOC is modeled as a function of OCV (fig. 4), a flat region occurs at SOC < 10%, OCV <3.4 volts.

The inventors model SOC as a function of OCV using Artificial Intelligence (AI), and performed the modeling only in region 1 (e.g., less than 5%) where the derivative is low and the error is also low. For example, as an example, region 1 may be selected with a derivative value of less than 5% or d (soc) < 0.5. Then, the inventors performed OCV test in zone 1 using the modeling result of zone 1. This greatly improves the accuracy of artificial intelligence modeling, modeling only region 1, but not regions 2 and 3.

The SOC, OCV data points obtained by the calibration process of fig. 9 may be plotted as points in a graph. The calibration curve is generated as a best fit function that best fits these data points. The calibration curve model generated by step 130 of fig. 9 is used by step 112 of fig. 10B to obtain modeled SOC values for 2 OCV voltage data points. The SOH is then calculated in step 118 of fig. 10B.

The calibration curve model can be obtained from AI modeling of these (SOC, OCV) data points, for example using least squares to find the parameters, using neural networks for optimization. Other statistical methods may also be used.

An Artificial Neural Network (ANN) may be used to generate an SOC model as an OCV function. Artificial neural networks are particularly useful for processing large amounts of non-linear data in complex ways that are difficult to define with traditional computer programs. Rather than being programmed with instructions, the training data is input to the neural network and compared to expected outputs, then adjusted within the neural network, and the training data is again processed and the outputs compared to produce further adjustments to the neural network. After a number of such training cycles, the neural network is changed to efficiently process data similar to the training data and expected output. Neural networks are an example of machine learning, as neural networks learn how to generate expected outputs for training data. Real data similar to the training data may then be input to the neural network to process the real-time data.

FIG. 7 shows a neural network used to model a calibration curve for SOC as a function of OCV. The input node 12 receives input data OCV, and the output node 60 outputs the result of operation of the neural network, SOC _ CALC, which is a modeled SOC value of the input OCV value. Two layers of operation are performed in this neural network. Nodes 20, 22, 24,. 28, 29 each take input from input node 12, perform wavelet function operations, and send outputs to nodes of the second layer. The second level nodes 52, 54,. 58, 59 also receive a plurality of inputs, combine the inputs to produce an output, for example by generating a product, and send the output to the third level nodes 60, which third level nodes 60 also combine or sum the inputs to produce an output.

The inputs to each stage are typically weighted, thus producing a weighted sum (or other weighted operation) at each node. Each input on a node may be assigned a weight that is multiplied by the input and then all weighted inputs are added, multiplied, or otherwise manipulated by the node to produce the output of the node. For nodes 20, 22, 24, … 28, 29 in the wavelet layer, these weights are designated A_ij、B_ijFor nodes 52, 54, … 58, 59 in the multiply layer, the weights are designated as W_ij. These weight values A_ij、B_ij、W_ijMay be adjusted during training. By trial and error or other training procedures or learning algorithms, paths that produce expected output may ultimately be given higher weight, while paths that do not produce expected output are given less weight. The machine learns which paths will produce the expected output and assigns high weights to the inputs on those paths.

These weights may be stored in weight memory 100 or in another memory. Since neural networks typically have many nodes, there may be many weights to store in the weight memory 100. Each weight may require multiple binary bits to represent the range of possible values for that weight. The weights typically require 8 to 16 bits. The weight memory 100 may be SRAM, DRAM, flash, disk, or various combinations of these or other computer storage devices.

Fig. 8 shows the training of the neural network using the measured SOC as a target to generate an old battery calibration model. In step 130 of fig. 9, the aged battery is measured, and the measured OCV and SOC data is stored and modeled. The measured OCV data is used as training data 34, OCV _ MEAS. The measured SOC data corresponding to the OCV _ MEAS value is recorded as target data 38, SOC _ MEAS. Each SOC _ MEAS value corresponds to the OCV _ MEAS value measured simultaneously over the aged battery life tested using the process of fig. 9.

The neural network 36 receives the training data 34 and the current set of weights A_ij、B_ij、W_ijAnd operates on the training data 34 to produce a result. The result of this generation is a modeled value of SOC (SOC _ CALC). The resulting result (SOC _ CALC) from the neural network 36 is compared to the target data 38(SOC _ MEAS) by a penalty function 42, the penalty function 42 producing a penalty value that is a function of how far the resulting result is from the target. The loss values generated by the loss function 42 are used to adjust the weights applied to the neural network 36. The loss function 42 may perform multiple weight iterations on the training data 34 until a minimum loss value is determined and the final set of weights is used for calibration curve modeling.

The neural network 36 may have multiple output nodes 60 to generate many SOC _ CALC values in parallel from parallel inputs of the OCV _ MEAS, rather than generating a single SOC _ CALC value. The loss function 42 may compare many SOC _ CALC values in parallel with many SOC _ MEAS values to generate a loss function value.

Fig. 9 shows a process of testing an old battery to obtain OCV and SOC values to model a calibration curve. The process of fig. 9 may be repeated with multiple old batteries to obtain a data set that may be input to a neural network (fig. 7, 8) to build a calibration curve model that may then be used to classify the old batteries (fig. 10A-10B, 11).

The old battery to be subjected to the calibration test is first discharged using a large constant current of 1C until the target minimum voltage Vmin is reached (step 121). After waiting for 3 minutes, the battery is further discharged at a constant current of 0.05C until the target minimum voltage Vmin is reached again (step 122). The cell was allowed to cool and rest for 24 hours.

After the rest period, the battery is charged with a Constant Current (CC) of 0.05C until the target maximum voltage Vmax is reached (step 124). The constant current is integrated over time to obtain Q. The battery tester can be used to record the battery voltage V of each time period or time step_chargeAnd an integrated current Q. Each of the plurality of data points has a Q and a V_chargeA value, stored or otherwise recorded.

The cell is discharged using a constant current of 0.05C until Vmin is reached (step 126). The 0.05C constant discharge current was integrated over time to obtain the current discharge capacity Q of the old cell_now. The battery tester can be used to record the battery voltage V of each time period or time step_dischargeAnd an integrated current Q. Each of the plurality of data points has a Q and a V_dischargeA value, stored or otherwise recorded.

V obtained during the charging step 124 for each of a number of Q values_chargeValue and V obtained during the discharging step 126_dischargeThe values are averaged to obtain an OCV value for the Q value (step 128).

The Q value is then converted to an SOC value (step 129). Since Q and SOC are linearly related to each other, for each data point, SOC can be calculated as SOC-Q/Q_now. Thus, (Q, V) obtained in step 124_charge) Data points and (Q, V) obtained in step 126_discharge) The data points have been converted to (SOC, OCV) data points.

The stored OCV and SOC data points are applied to an artificial intelligence engine (fig. 7, 8) to generate an SOC model, i.e., a calibration curve, as a function of OCV (step 130).

Fig. 10A-10B are methods of testing and sorting used batteries based on two (Q, OCV) data points measured during a short-zone OCV test. The voltage of each old battery is measured as Vcel (step 102). When Vcel is higher than the maximum voltage Vmax or lower than the minimum voltage Vmin (step 104), the battery is disposed of (step 106). The battery, with initial voltage Vcel between Vmin and Vmax (step 104), is further processed.

The battery voltage Vcel is compared to the OVC voltage range of region 1 (step 108). For the cell of fig. 6, region 1 is OCV <3.4 volts. When Vcel is greater than 3.4 volts (step 108), the battery is discharged for a short period of time (step 110), after a period of rest, the battery voltage Vcel is again compared to the OCV voltage range of zone 1 (step 108). The discharging and comparing of the cell is repeated until its voltage falls within region 1 (step 108).

In fig. 10B, once the battery voltage is within the target OCV region, a regional OCV test is performed on the old battery (step 134). The battery tester discharges the battery between two voltages OCV1 and OCV 2. While the discharge current deltaq is measured. The OCVs 1 and the OCV2 are close to each other, for example, when the OCV1 is equal to the OCV2+0.1 volts, and thus the charge/discharge current Δ Q and the charge time are small. The OCV1 may be the latest voltage of the battery, i.e., Vcel measured in step 108, although the battery tester may measure the battery voltage again. This test is shown in more detail in figure 11.

The calibration curve modeled by AI in step 130 of fig. 9 is used. The OCV1 is input to a calibration curve model that outputs the SOC1, and then the OCV2 is input to the calibration curve model to obtain the SOC 2.

At step 118, the state of health (SOH) is determined from SOC1, SOC2, Δ Q, and Q using the following equations_newCalculating:

SOC1 and SOC2 are obtained from the AI calibration curve model at step 112, and Δ Q is obtained from coulomb counting, Q at step 134_newIs the full charge capacity of the new battery specified by the manufacturer or a new battery may be measured.

The SOH of the battery under test is compared to a SOH threshold, e.g., 75% (step 114), and batteries with SOH below the threshold are disposed of (step 132). Batteries above the SOH threshold are sorted into quality bins according to their SOH values (step 116). The sorted batteries may be reused for various applications according to quality sorting. Some applications may require higher quality recycled batteries than other applications. For example, batteries with SOH higher than 95% are more expensive and may be used for more demanding applications than batteries with SOH between 80% and 75%.

Fig. 11 shows the regional OCV test in more detail. The battery tester records the battery voltage Vcel as OCV1 (step 140). The old battery to be tested was discharged using a constant current of 0.05C until a target voltage approaching OCV2 was reached (step 142).

This test is rapid because OCV2 is very close to OCV1, e.g., OCV2 ═ OCV 1-0.1 volts. The battery tester may continuously test for a voltage drop of 0.1 volts or may discharge for a short fixed period of time (e.g., 8 minutes).

The tester also measures the discharge current Δ Q provided by the battery as the battery voltage drops from OCV1 to OCV2 (step 142).

The battery was allowed to cool and rest for one hour before the OCV2 was read by the battery tester (step 144). The early reading of OCV2 by the battery tester is not accurate because the battery has not been at rest. Due to time effects, the final difference between OCV1 and OCV2 may not be exactly the target voltage drop, such as 0.1 volts.

After rest, OCV2 is saved in computer memory or otherwise recorded along with Δ Q and OCV1, which may have been recorded in advance. These values may be stored by writing to a computer memory, such as a register file, SRAM, DRAM, or hard disk.

Alternative embodiments

Several other embodiments are also contemplated by the inventors. For example, for certain battery chemistries or battery packs having multiple cells, the region with the lowest derivative value may not be the region with the lowest OCV voltage (fig. 6). Step 108 of fig. 10A may compare the battery voltage Vcel to the upper and lower limits of OCV in this lowest derivative region, and the battery may be charged to increase Vcel until it is within this region. Step 110 of fig. 10A may charge the battery for a short time when Vcel is less than the OCV lower voltage limit of the region, or discharge the battery for a short time when Vcel is greater than the OCV upper voltage limit of the region. The charging and discharging time in step 110 may be adjusted by the difference between Vcel and the nearest boundary of the target area, for example, increasing the charging time when Vcel is further away.

The OCV region having the lowest derivative value may be selected by comparing the derivative value with a threshold value (e.g., 0.5) to determine upper and lower limits of the OCV of the region. There may be multiple regions, or there may be multiple discrete portions of the region, requiring more comparisons to be made for multiple boundaries of the region.

In fig. 9, the initial step 121 may be deleted. Step 121 may reduce test time by discharging the battery more quickly at a high current. This is particularly useful when the target zone has the lowest OCV, as the cells of the target zone are near a fully discharged state.

While fig. 11 depicts discharging the battery from OCV1 to OCV2, the battery tester may also charge the battery from OCV1 to OCV 2. The polarity of the values in the calculation may be adjusted for charging rather than discharging, or absolute values may be used.

The order or sequence of some steps may be changed. As one example, storing the OCV and SOC data may occur during step 129 or may be performed over several steps 124 and 129. Various modifications to the neural network are possible, such as more layers or weights or different functions. More sample points may be input and more iteration cycles or periods may be used. Neural network modeling and optimization can be used to obtain a very good fit to the calibration curve model.

The calibration curve model may be implemented as a lookup table that outputs a modeled SOC value when the measured OCV is input to the lookup table. The calibration curve may also be implemented as a function performed by a processor (e.g., a microprocessor, central processing unit, arithmetic logic unit, co-processor, or other programming machine). The memories may be shared or separate, local, remote, or various combinations, and the processors and other computing blocks may be shared, distributed, local, remote, or various combinations.

The calibration endpoint may be based on a SOC threshold, or the collection of OCV and SOC data may be stopped after a certain number of data points have been collected, or after a certain amount of time has elapsed, or some other criteria. The test technician may simply run out of time and stop further data collection, and then proceed to generate a model for the calibration curve. An initial model may be generated for use and then a finer model generated from more data points.

Although integration of current to produce Q has been described_nowAnd Δ Q, but for the integration of the constant current, it may be the constant current multiplied by the time period during which the constant current is applied. Various approximate integration methods may be applied, such as using PWL or multiplying the current by each of several short time periods. Coulomb counting methods can be used to integrate charge over time. The integration method may accumulate the charge transferred over a small period of time.

Although an initial deep discharge is not required, the battery may be pre-discharged or pre-charged in other steps, if desired. The rest period can be shortened or lengthened. A simple battery stand test apparatus may be used rather than a complex test stand.

The calibration curve may be approximated by one or more functions, such as Piece-Wise Linear (PWL) or multivariate functions. SOC can be modeled with equations for terms such as square root, logarithm, etc. of OCV.

The temperature of the cell should be maintained at a constant value, such as room temperature, during the test. The length of the rest period after charging or discharging the battery may depend on the charging/discharging current and the thermal performance of the battery. The thermal performance of a battery may change with age, for example, due to increased internal resistance, resulting in increased heating of the old battery.

Many parameters and values may be altered from the given example. The voltages, e.g., Vmax, Vmin, OCV1, OCV2, etc., and the current C may have different values, or different ratios from each other. Imin may be 0.05C, Vmax may be 4.2 volts, and Vmin may be 2.75 volts, to name just one of many examples. Rather than using an old battery for calibration, a new battery may go through many charge/discharge cycles to induce aging.

The number of cells tested for calibration may be a relatively small number, e.g., 3 cells may be tested when artificial intelligence modeling is effective, or more cells, e.g., 100 cells, may be tested when less effective modeling is used, or when more accurate calibration is required. Some battery reuse applications may not require accurate SOC modeling. Ideally, the batteries tested for calibration are closely related to the batteries being screened, e.g., the same manufacturer and model. The battery under test may be a single battery or a battery pack, a single battery or a plurality of batteries.

Some test errors may be tolerated depending on the application or intended use of the recycled battery. In some cases, a test error of +/-3% of the actual SOH may occur. When larger currents can be used to achieve the required test accuracy or error tolerance, the test time may be reduced.

The current may be positive or negative, and terms such as charging and discharging may be used interchangeably depending on the polarity of the current. While a constant current has been described, a variable current may be used and integrated over time to obtain the Q value.

Some embodiments may not use all components. Other components may be added. The penalty function 42 may use various error/penalty and cost generators, such as weight decay terms, to prevent the weights from growing too much over multiple cycles of training optimization; sparsity penalties may then encourage nodes to zero their weights, so only a small fraction of the total nodes do so. Many alternatives, combinations, and variations are possible. Other variation and loss or cost terms may be added to the loss function 42. The values of the relative scaling factors of the different cost functions may be adjusted to balance the effects of the various functions. The training endpoint of the neural network may be set for a combination of conditions, such as a desired final precision, precision-hardware cost, target hardware cost, and the like.

The neural network 36, the penalty function 42, and other components may be implemented in a variety of technologies using various combinations of software, hardware, firmware, routines, modules, functions, and so on. The final result, i.e., the calibration curve model or calibration function generator, may be derived from the neural network 36 with the final weights and may be implemented as program modules or in an Application Specific Integrated Circuit (ASIC) or other hardware to increase processing speed and reduce power consumption.

The background of the invention may contain background information related to the problem or environment of the invention rather than the prior art described in connection with others. Accordingly, the inclusion of material in the background section is not an admission of prior art by the applicant.

Any methods or processes described herein are machine-implemented or computer-implemented and are intended to be performed by a machine, computer, or other device, and not merely by a human being without machine assistance. The tangible results produced may include reports or other machine-generated displays on display devices such as computer displays, projection devices, audio generation devices, and related media devices, and may include hardcopy printouts that are also machine-generated. Computer control of other machines is another tangible result.

Any advantages and benefits described herein are not necessarily applicable to all embodiments of the invention. When the term "means" appears in a claim element, it is the intention of the applicant that such claim element fall within the definition of 35USC, section 112, clause 6. Typically, the word "device" is preceded by one or more words of labels. The word or words preceding the word "means" is a label intended to facilitate the recitation of claim elements, and not intended to convey a structural limitation. The means-plus-function claims are intended to cover not only the structures described herein for performing the function and their structural equivalents, but also equivalent structures. For example, although a nail and a screw have different configurations, they are equivalent structures in that they both perform the fastening function. Claims that do not use the term "device" do not fall within 35USC, section 112, clause 6. The signals are typically electronic signals, but may also be optical signals, which may be transmitted, for example, by fiber optic lines.

The foregoing descriptions of embodiments of the present invention have been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. It is intended that the scope of the invention be limited not by this detailed description, but rather by the claims appended hereto.