Marine Engine Fault

This record contains an experimental dataset for condition monitoring, fault diagnosis, and anomaly detection on a marine diesel engine. All data were acquired on a Matsui Iron Works MU323DGSC engine under controlled test-bench conditions. The dataset comprises a baseline reference-performance recording together with five induced fault/anomaly classes, each recorded at one or more fixed engine loads, providing labelled time-series suitable for developing and benchmarking data-driven diagnostic and anomaly-detection methods.

The five fault scenarios are:

• Compressor air-filter clogging — 40, 60, 75, 85 % load
• Air-cooler fouling — 40, 60, 75, 85 % load
• Injection-valve nozzle clogging — a one-hole restricted nozzle (40 / 60 / 85 % load) and a more severe two-hole restricted nozzle under a load program
• Cooling-water pump cavitation — 60, 85 % load
• Turbine degradation — 40, 60, 85 % load

Each scenario was produced by physically inducing the corresponding abnormal condition on the engine; the procedure for each is described in the README.

Data records. The dataset contains 16 CSV files — one baseline reference file and 15 fault-scenario files — totalling roughly 115,000 time-stamped samples across about 70 measured and derived channels (engine speed, in-cylinder pressures, exhaust and cooling-circuit temperatures, flows, pressures, and work/efficiency quantities). All files are UTF-8 encoded and use a consistent three-row header (full variable name, shorthand symbol, unit). The scenario files share a single 73-column schema whose first three columns are Time_abs, Time_rel, and a binary Anomaly State label (0 = pre-anomaly, 1 = anomaly); the two injection-nozzle files contain anomalous operation only. The reference file uses a separate 70-column baseline schema (no anomaly label). Two channels (Compressor Filter Loss and Turbine Back Pressure) were not recorded in five runs and are left empty in those files. A file-level index (dataset_index) and a variable dictionary (variable_dictionary) documenting every channel, its unit, data type, and per-file presence are included, along with the script used to normalise the raw exports.

Associated publication. This dataset accompanies the Data Descriptor "[manuscript title]" ([author list]), submitted to Scientific Data; a preprint is available at arXiv:[ID]. Please cite both the Data Descriptor and this dataset when using the data.

License. Creative Commons Attribution 4.0 International (CC BY 4.0).