Early detection of anomalous operation of battery systems is critical in improving performance and ensuring safety. This paper presents an effective and computationally efficient approach for online anomaly detection using real-time voltage and temperature data from multiple Li-ion cells. Median-based residuals are generated and evaluated using cumulative sum control chart to detect anomalies. Due to the scarcity of anomalous data for evaluating the anomaly detection algorithm, we inject anomalies into nominal experimental data using a physics model-based approach. The proposed anomaly detection approach has low false positive rate and has the capability to detect significant voltage and temperature anomalies. The approach accurately detects anomalies with voltage and temperature deviations greater than 15mV and 1.3°C, respectively, without any missed detection. The approach accurately traces the anomalous cells, distinguishes voltage and temperature anomalies, and identifies the direction of detected anomalies.