Data Pre-processing

Data Pre-processing Flowchart  

Figure 1: Flowchart representing the steps taken for data pre-processing

From the flowchart above, the first step of obtaining the dataset and getting it verified was obviously already done. The second step of picking out irregularities in the dataset involved loading it into Matlab and basically observing the patterns in the data and any discontinuity, errors, missing information and unsupported data formats that Matlab would not be able to detect. 

The only issues found in the dataset was the format in which time was given, absence of certain variables and unexplained variables. These missing and unformatted data were effectively and easily picked out without using any exploratory tools but rather just by using knowledge on supercapacitors. The first run made was that it was seen that when the charge and discharge curves were plotted using the given time values, the graph was seen to be quite staggered. This was a result of matlab not recognizing the time format given in the dataset. The hour, minutes and seconds were all converted to seconds where by then it was able to be imported into matlab. In addition, another issue found with the given time was that the time for the charging half started from zero and as soon as the discharging half started the clock was reset to 0. Using this time format would be impractical because for visualization purposes, the graphs would return to zero every discharging half, as such, the group had to edit the dataset and manually enter the time so that the time started from the charging half and continued to the discharging half until the last cycle. In addition, there was a certain variable that was recorded in the dataset but had no justification of how it was computed and this was the charge and discharge energy. These variables were attempted to be calculated based on the formulas stated in paper [1], however were seen not to match. This suggested that the values were instantaneous meaning when the testing was being done for each capacitor, its energy was logged for that particular time.

......Tired of reading yet??? 

 

Lets watch this instead, garb your snacks.........

Exploring the Dataset

https://youtu.be/MoOp4ULuXZM

OR, scan the barcode below

Computing the Capacitance

The first variable that was calculated was the capacitance value. The formula for calculating the capacitance is denoted by:

The instantaneous current, It, was computed as discharging current multiplied by the discharging time while the change in voltage, was computed by subtracting the end voltage with the initial voltage. An important point to note is that the computations for the capacitance was solely reliant on the discharging phase of the capacitor because it is important to know how much energy the device can deliver rather than the amount it can store.

Now we visualize the typical degradation of a supercapacitor with respect to the drop in its capacitance value over time of usage

Figure 2: Capacitance vs time graph

The graph shown above illustrates the drop in capacitance overtime for the capacitors tested in batch 4 of the dataset. According to the datasheet of the HV, series 1 F 2.7 V commercial grade supercapacitors, the capacitor would decrease in overall capacitance by 30% after 500000 cycles but since the dataset only contained 10000 cycles worth of data, a threshold could be manually set. For example, if the capacitor has a capacitance of 1F than if the threshold is 93%, that means the capacitor would be considered to have failed once it reaches 0.93F and below. 

Pre-recorded Variables 

The dataset contained several pre-recorded features that could be used to predict the capacitances. However before doing that, it is very important to observe and understand how one variable is related to the other to fully grasp how the change in one affects the change in the other. This is what the model will look at once it is fed with the predictors. The system will anticipate the type of changes that the output variable will experience and will accordingly predict the outcome based on these factors.

Some of these pre-recorded variables were: 

• Cycle: Cycle: Cycle represents and explains the operation of the supercapacitor throughout its discharging and charging phases. From the capacitance graph it was observed that as the cycle increases throughout a capacitor’s life cycle, the capacitive output of the device decreases thus this feature was deemed to be the most influential.

• Charge Capacity: the amount of current that can be stored by a supercapacitor at an instance of time is called the charge capacity and is given in mAh. According to [2], the current holding capacity of the supercapacitor decreases with time due to various factors such as aging which refers to the electrolyte and electrode materials undergoing chemical reactions. Another factor is electrolyte loss that refers to the internal gel or liquid electrolyte bases gradually evaporating or seeping out overtime which alternately causes the device to lose its ability to store the same amount of charge than it used to.

• Discharge Capacity: juts like the charge capacity, discharge capacity refers to the amount of current that the device can deliver. This feature represents the ability of the supercapacitor to provide constant current over a period of time. Initially, the mAh rating of a capacitor will deviate with long cycles of usage due to leakage current. As discussed in [3], leakage current arises from small amounts of internal losses due to mechanical stress or electrode degradation. This causes the stored current to slowly self-discharge over a period of time which leads to lower levels of current left for discharge to the load which ultimately contributes to the decrease in discharge capacity.

• Charge Energy: Charge energy is typically measured in Wh or mWh in the case for the utilized supercapacitor. This feature represents the amount of energy that a supercapacitor can hold and made available for discharge when required at a given voltage. This parameter degrades overtime due to cycling and aging whereby the device undergoes change in temperature leading to loss of integrity of the chemical and mechanical components. In addition to that, all supercapacitors have internal resistance which increases overtime and leads to loss of the energy storage capacity and reduction in charge retention [3].

• Discharge Energy: Similar to charge energy, discharge energy refers to the amount of energy that can be delivered from the stored energy in mWh. Like the losses arising from discharge capacity, the increase in internal resistance and degradation of electrode structure over many cycles of use can cause the amount of energy being delivered to significantly drop [5].

In the next post, we are going to gaze into some of the engineered features that are going to be included into the feature set.... Stay tuned. Thank you

References

[1] J. Ren et al., "Engineering early prediction of supercapacitors’ cycle life using neural networks,"                     Materials Today Energy, vol. 18, p. 100537, 2020.

[2] C. Liu, Q. Li, and K. Wang, "State-of-charge estimation and remaining useful life prediction of                     supercapacitors," Renewable and Sustainable Energy Reviews, vol. 150, p. 111408, 2021 

[3] M. Yassine and D. Fabris, "Performance of commercially available supercapacitors," Energies,                     vol. 10, no. 9, p. 1340, 2017.

[4] Z. S. Iro, C. Subramani, and S. Dash, "A brief review on electrode materials for supercapacitor," Int.                J. Electrochem. Sci, vol. 11, no. 12, pp. 10628-10643,2016.

Edited by Shahil and Henal

S11172483@student.usp.ac.fj

S11085370@student.usp.ac.fj