Dataset generators for training and validation of classification models for fault diagnosis based on artificial neural networks
Software generators of the data for training and validation of the neural networks for vibrodiagnostics of rotating machines comprise three separate solutions for generating training and validation datasets.
The first generator is created in the MATLAB environment and generates vibration acceleration signals, which consist of a defined number of harmonic components with random frequency, amplitude and random amount of additive noise. Software generates two classes of signals with different bandwidth. The first class is limited by a 1st order filter with a cutoff frequency of 1 kHz, the second with a cutoff frequency of 11 kHz. Generated signals are exported to a * .CSV file and can be further used for training and validation of the neural network.
The second generator is implemented in the LabVIEW environment and uses the * .CSV output from the first generator and transforms a short section of the generated signal into an infinite length signal. This is then sent to a vibration exciter, on which a smart diagnostic sensor is located, which senses mechanical signals with two accelerometers. The generator communicates with the sensor via the digital bus and stores the data captured by sensors with different bandwidths. This real data reflecting actual mechanical properties of the sensing system is then used as the input data for neural network training and validation.
The third generator is also implemented in the LabVIEW environment and generates a clearly user-defined frequency spectrum with additive noise and a given bandwidth using a low-pass 1st order filter with a user-defined cutoff frequency. The generator output is *. CSV file, which can be imported e.g. into the Python environment for training and validation of the neural network.