Machine learning for computationally efficient electrical loads estimation in consumer washing machines

Abstract Estimating the wear of single electrical parts a home appliance without resorting to large number sensors is desirable for ensuring proper level maintenance by manufacturers. Deep learning techniques can be effective tools such estimation from relatively poor measurements, but their computational demands must carefully considered, actual deployment. In this work, we employ one-dimensional Convolutional Neural Networks and Long Short-Term Memory networks infer status some components different models washing machines, signals measured at plug. These are trained tested on dataset (502 cycles $$\approx$$ ≈ 1000 h) collected four machines designed in order comply with memory constraints imposed available hardware selected real implementation. The approach end-to-end; i.e., it does not require any feature extraction, except harmonic decomposition signals, thus easily generalized other appliances.