Tohoku Dataset

The second dataset used in this study is contributed by students at Tohoku University. In this dataset, we are interested in detecting capacity degradation anomalies in the cycling data of solid-state lithium-ion batteries. For example, in the figure below, we can observe a sudden drop in the discharge capacity at the cycle index 79, 429 and 476, which indicates an anomaly in the battery’s performance. This capacity degradation anomaly is an example of point anomaly that we aim to detect using our proposed OSBAD framework.

The dataset consists of cycling data from 10 cells, each identified by a unique cell index. The cycling data includes measurements such as voltage and discharge_capacity over multiple charge-discharge cycles.

Battery Chemistry Description

Property	Tohoku
Positive electrode	NMC523 (LiNi₀.₅Co₀.₂Mn₀.₃O₂)
Electrolyte	Solid electrolyte (Li₆PS₅Cl)
Negative electrode	In/InLi
Number of cells	10 cells
Nominal capacity	100 mAh/g
Discharging C-rates	0.1 C (1C ≈ 233 μA)
ΔSOC in %	100%
Theoretical voltage limits	3.0 V - 4.3 V vs. Li⁺/Li
Operating temperature	25 °C
Anomalous data type	Discharge-capacity profile
Data source	This paper

Feature Description

Feature	Type	Description	Range/Values	Anomaly Relevance
discharge_capacity	Float	Discharge capacity measured during the discharge phase (mAh)	0.0 - 105.66	High - Primary indicator of capacity fade anomalies
cycle_index	Integer	Sequential charge-discharge cycle number	0 - 499	High - Temporal reference for anomaly detection
voltage	Float	Discharge voltage measured during cycling (V)	1.25 - 3.63	Medium - Voltage measurement including anomalies.
cell_index	String	Unique identifier for each battery cell	Cell-specific labels	Low - Grouping variable for per-cell analysis
outlier	Integer	Ground-truth anomaly label (0=normal, 1=anomalous)	0, 1	High - True label for validation and evaluation

Important

The discharge_capacity is the primary feature of interest for detecting capacity degradation anomalies, while cycle_index provides the temporal context. The voltage feature can also provide insights into battery health, but is secondary to capacity measurements. The cell_index allows for analysis on a per-cell basis, and the outlier label is used for validating the performance of anomaly detection methods.