Issue link: https://resources.mouser.com/i/1481902
Intelligence at the Edge 14 AI Development Pipeline As a rule, these are the six steps for adding machine learning to edge devices so they can learn and infer independently (Figure 1) NanoEdge™ AI Studio eases this development process by automating the data science and algorithm related steps. DATA ACQUISITION Once a well-defined use case with a specific set of monitoring conditions exists, the first development step is data collection for ML algorithm selection and training. This step is conceptually simple but takes time. Data is acquired for each class of machine states so the ML algorithm can identify commonalities between data points of the same class as well as the differences with datapoints of other classes. For the brushless motor classification algorithm example, this means collecting current data of the motor running nominally under as many diverse loads and conditions as possible, while also collecting current data on each of the abnormal behavior classes—either by extracting the signal information from existing data sets or by recreating each failure mode to generate the data. Data collection for anomaly detection is similar. Signals are acquired for the motor working nominally under different conditions, as well as for example anomalies that we want the algorithm to predict. DATA PROCESSING Data processing is the transformation of raw data into a useful and efficient format. This transformation can be as simple as normalization or it may involve feature creation and extraction such as running a Fast Fourier Transform (FFT). This step is often associated with ML algorithm searching and training, but it is an opportunity for engineers to exploit their use case knowledge to improve efficiency. If, for example, relevant information is being monitored in the frequency domain for a particular use case, an engineer can implement the known FFT. Otherwise, different possible combinations of processing steps may be needed to obtain the best results in the next step. NanoEdge™ AI Studio automatically searches all processing combinations available for the Arm® Cortex® cores of STM32 MCUs and STM32MP1 MPUs. Many robust ML algorithms are available for their Arm® Cortex®-Mx and -A7 CPU cores. Figure 1: Machine Learning development steps. (Source: STMicroelectronics)