Machine learning and artificial intelligence models are increasingly common in predictive maintenance due to their ability to automatically extract high-level features with less human intervention. These models have been shown to give good results in machinery or rotatory fault diagnosis. However, due to the complexity of vibration and audio signals used in fault diagnosis, some pre-processing is required before feeding the machine learning algorithm. Fast Fourier Transform (FFT) and the Hilbert transform (HT) envelope spectrum are mostly used in the literature. However, these frequency domain transforms are not very effective when applied to rotating systems (bearings) fault detection. Indeed, in these applications, fault signal patterns are usually very weak relative to background noise and other interference in the early damage stage. In this paper we propose to use signature to extract new features from sensor data and use these new feature to train machine learning models. The main idea is to use the extracted signature coefficients to create 2d images that are then fed to a deep neural network for classification. Our experimental results show that this method outperforms most state-of-the-art methods on eight (8) bearing fault diagnosis data sets and three (3) other time series classification data sets. For example, on the Case Western Reserve University (CWRU) data set, the proposed method accuracy ranges from 96.59 % to 100% accuracy. Moreover, the results show that this method is particularly well suited for high dimensional time series. The results also show that compared to Fast Fourier Transform (FFT), the signature method requires fewer data points to detect failure. This means that in a situation where the two methods have similar performances, the signature method detects failure faster than FFT. This is very important in fault detection and predictive maintenance, in particular where it is crucial to detect faults before they occur or get worse.