ESPset

A real-world dataset for vibration-based fault diagnosis of electric submersible pumps used on offshore oil exploration. The ESPset dataset is a collection of vibration signals acquired from accelerometers strategically attached onto the components of Electrical Submersible Centrifugal Pumps (ESP). An ESP belong to a class of equipment used in the extraction and exploration of oil and gas subject to severe working conditions. An ESP system consists of a coupled set of one or more electric motors, pumps and protectors.

# spectrum.csv
This csv file is a matrix of 6032 lines and 12103 columns, whose values are float numbers separated by a ';'. Each line of this file contains the spectrum of a single vibration signal collected from a sensor at a specific test condition of the ESP. Each value is the amplitude in inches per second (velocity) at a specific frequency. Each signal is normalized by the rotation frequency at which the ESP operates, in such a way that the amplitude with respect to the rotation frequency is always at the same position for all signal arrays. For instance, the normalization is made such that the 3003rd position of the array is always the 1x harmonic. Consequently, the 6006th is the second harmonic (2X). Hence, the frequency resolution is `1/3003~0.000333`

# features.csv
This csv file of 6033 lines (one line for each signal + a header) contains some features and the labels for all signals:
- esp_id: The id of the ESP.
- label: The classification label.
Let X be defined as the rotation frequency of the ESP. Each feature is defined as:

- median(8,13): Median of the amplitudes in the interval (8% X, 13% X);
- median(98,102): Median of the amplitudes in the interval (98% X, 102% X);
- a: Coefficient a of the exponential regression of type e^(a*A+b) where A is an array of equally separated relative frequencies up to 0.4, excluding zero. Example: A=(0.01, 0.02, ..., 0.39, 0.4).
- b: Coefficient b of the exponential regression of type e^(a*A+b) where A is an array of equally separated relative frequencies up to 0.4, excluding zero. Example: A=(0.01, 0.02, ..., 0.39, 0.4).
- peak1x: Amplitude in X;
- peak2x: Amplitude in 2X;
- rms(98,102): Root mean square of the amplitudes in the interval (98% X, 102% X).

For more information and code, take a look at https://github.com/NINFA-UFES/ESPset