# Lithium battery health diagnosis and remaining life prediction-Part 1

**Background introduction**

**Introduction to accelerated aging experiment**

**1 Experimental equipment**

**Table 1 Accelerated aging experiment scheme**

**Core algorithm introduction**

The principle of using the IC curve to estimate the SOH of a battery is that when the SOH of the battery is different, the change in the capacity corresponding to the same value of the battery voltage change is different. Based on the same principle, the author of this article extracted 6 different voltage ranges (3.6~3.7 V, 3.7~3.8 V, 3.8~3.9 V, 3.9~4.0 V, 4.0~4.1) from the charging voltage curve at 0.5C of the battery. V and 4.1~4.2 V) corresponding to the capacity difference as the characteristic of the estimated SOH, as shown in Figure 3.

Figure 3 Feature extraction of battery cell attenuation

It can be seen that as the battery SOH decreases, the extracted cumulative capacity value also decreases, so it can be used as a health indicator. Finally, the relationship between health indicators and battery SOH is constructed through the method of machine learning. Six health indicators (HI1~HI6) are used as the input layer of the neural network, and then fully connected with 50 hidden layer neurons, and finally output the battery capacity to obtain SOH.

Figure 4 Neural network structure

2. The second step of the algorithm is to identify the battery aging stage. Figure 5 shows the attenuation curves of battery cells at different discharge rates. Each capacity attenuation curve can be divided into two stages, the first stage is the first 100 cycles, and the attenuation is relatively fast; the second stage attenuation curve is almost linear and relatively flat. The two aging stages can be divided by curve fitting tools; when the fitted straight line R-square is the smallest after division, it is the optimal division position. The author of this article also uses the capacity decay curve under 0.5C to identify the aging stage of the battery, and then uses the random forest (RF) algorithm, using the six health indicators obtained in the first step of the algorithm as input, and constructs health with 100 trees The relationship between the indicator and the aging stage, and finally the aging stage of the output battery.

Figure 5 The decay curve of single battery capacity under different discharge rates

>>>TO BE CONTINUED...