The CAFFEINE dataset contains non-intrusive sensor data (time series) and labels (coffee type for each time series, and actuator activation status at each time step) for 130 coffees. Eight sensors are placed along the power chain of the coffee making process (1 current sensor, 1 voltage sensor, 3 accelerometers, 2 temperature sensors, 1 coffee level sensor), producing signals originated by 5 sources (heating coil, infuser translation motor, grinder, vibration pump, (electronics)), sampled at 6250 Hz. The dataset comes with reading scripts and instructions for Python and MATLAB users. Intended uses for this dataset include blind source separation (multi-label clustering and signal decomposition), classification, as well as regression (multivariate time series forecasting), parameter identification and model synthesis.