Industrial ML Datasets

The following is a curated list of datasets, publically available for machine learning research in the area of manufacturing.

For more information, please check our corresponding publication:

@inproceedings{jourdan_machine_2021,
 title = {Machine {Learning} for {Intelligent} {Maintenance} and {Quality} {Control}: {A} {Review} of {Existing} {Datasets} and {Corresponding} {Use} {Cases}},
 volume = {2},
 journal = {Proceedings of the 2nd Conference on Production Systems and Logistics},
 author = {Jourdan, Nicolas and Longard, Lukas and Biegel, Tobias and Metternich, Joachim},
 year = {2021},
}

Some additional datasets may be found here: Link

✔️ indicates a preset split between training and testing data.

🌐 indicates, that the test set labels are hidden behind an evaluation server.

Predictive Maintenance and Condition Monitoring

Name	Year	Feature Type	Feature Count	Target Variable	Instances	Official Train/Test Split	Data Source	Format	License	Access
Wood veneers before and after drying This dataset consists of 2579 image pairs (5158 images in total) of wood veneers before and after drying.	2021	Image	>4000x4000	-	5.158	❌	Real	PNG	CC BY 4.0	Link
Diesel Engine Faults Features Fault detection based on pressure curves and vibration.	2020	Signal	84	C (4)	3.500	❌	Synthetic	MAT	CC BY 4.0	Link
Degradation of a Cutting Blade Wrapping machine process data over 12 months with a degrading cutting tool.	2019	Signal	9	C (8) / R	1.062.912	❌	Real	CSV	CC BY-SA 3.0	Link
CNC Mill Tool Wear CNC process data of wax milling with worn/unworn tools.	2018	Signal	48	C (3*2)	25.286	❌	Real	CSV	CC0: Public Domain	Link
Condition Monitoring of Hydraulic Systems Test rig process data of multiple load cycles with various fault types and severity levels.	2018	Signal	17	C (5*(24))	2.205	❌	Real	Non-Standard	?	Link
Production Plant Data for Condition Monitoring nonymized process data of component run-to-failure experiments.	2018	Signal	26	-	228.414	❌	Real	CSV	CC BY-SA 3.0	Link
Versatile Production Popcorn production process data with multiple process steps.	2018	Signal	5-85	-	80.000	❌	Real	CSV	CC BY-NC-SA 4.0	Link
Degradation Measurement of Robot Arm Position Accuracy Target- and actual values of robotic arm tool position, velocity and current for health assessment.	2017	Signal	73	-	155.000	❌	Real	CSV	?	Link
APS Failure at Scania Trucks Anonymized counters and histograms for air pressure system fault detection.	2016	Signal	170	C (2)	76.000	✔️	Real	CSV	GNU General Public License	Link
Maintenance of Naval Propulsion Plants Gas turbine process data for component decay state prediction.	2016	Signal	16	R	11.934	❌	Synthetic	Non-Standard	More Information Use of this dataset in publications must be acknowledged by referencing the following publication: A. Coraddu, L. Oneto, A. Ghio, S. Savio, D. Anguita, M. Figari, Machine Learning Approaches for Improving Condition?Based Maintenance of Naval Propulsion Plants, Journal of Engineering for the Maritime Environment, 2014, DOI: 10.1177/1475090214540874, (In Press)	Link
Plant Fault Detection Anonymized process data for plant fault detection.	2015	Signal	10	C (6)	8.938.370	❌	Real	CSV	?	Link
Asset Failure and Replacement Anonymized data for asset fault detection.	2014	Signal	1	C (2)	447.341	✔️ 🌐	Real	CSV	?	Link
Maintenance Action Recommendation Anonymized process and maintenance data of an industrial asset for maintenance action recommendation.	2013	Signal	32	C (14)	2.097.152	✔️ 🌐	Real	CSV	?	Link
Anemometer Fault Detection Anemometer measurements for fault detection.	2011	Signal	16 16-20	-	345.700 208.800	✔️ 🌐	Real	Non-Standard	?	Link
Gearbox Fault Detection Test rig accelerometer data for fault detection.	2009	Signal	3	-	> 10 Mio.	❌	Real	CSV	?	Link
Li-Ion Battery Aging Battery test rig data during charge and discharge cycles for degradation detection.	2008	Signal	12	-	2.167	❌	Real	MAT	N/A	Link
Turbofan Engine Degradation Simulation C-MAPSS simulation sensor data of various conditions and fault modes.	2008	Signal	26	-	262.256	✔️	Synthetic	Non-Standard	?	Link
Bearing Bearing test rig accelerometer data of run-to-failure experiments.	2007	Signal	4-8	-	61.440	❌	Real	CSV	?	Link
Milling Milling process- and external sensor data for tool wear detection.	2007	Signal	13	R	1.503.000	❌	Real	MAT	?	Link
CWRU Bearing Data Bearing test rig accelerometer data for fault detection.	n.A.	Signal	5	C (2)	> 10 Mio.	❌	Real	MAT	?	Link

Process Monitoring

Name	Year	Feature Type	Feature Count	Target Variable	Instances	Official Train/Test Split	Data Source	Format	License	Access
Skoltech Anomaly Benchmark (SKAB) Time-series data from water circulation loop testbed for evaluating anomaly detection algorithms.	2020	Signal	8	C (2)	34×1,200	✔️	Real	CSV	GNU GPL v3.0	Link
High Storage System Anomaly Detection Storage test rig process data for anomaly detection.	2018	Signal	20	C (2)	91.000	❌	Synthetic	CSV	CC-BY-NC-SA 4.0	Link
Genesis Pick-and-Place Demonstrator Material sorting test rig process data for anomaly detection.	2018	Signal	23	C (3)	32.440	❌	Real	CSV	CC-BY-NC-SA 4.0	Link
Tennessee Eastman Process Simulation Dataset Simulated chemical process data for anomaly detection with different fault types.	2017	Signal	51	C (21) / R	> 10 Mio.	✔️	Synthetic	RData	More Information The person who owns, created, or contributed a work to the data or work provided here dedicated the work to the public domain and has waived his or her rights to the work worldwide under copyright law. You can copy, modify, distribute, and perform the work, for any lawful purpose, without asking permission.	Link
Robot Execution Failures Force and torque measurements of an industrial robot with different erroneous operating conditions.	1999	Signal	89	C (13)	463	❌	Real	Non-Standard	?	Link
Mechanical Analysis Vibration measurements of electromechanical devices with different erroneous operating conditions.	1990	Signal	7	C (6)	209	✔️	Real	MAT	?	Link
CWRU Bearing Data Bearing test rig accelerometer data for anomaly detection.	n.A.	Signal	5	C (2)	> 10 Mio.	❌	Real	MAT	?	Link

Predictive Quality and Quality Inspection

Name	Year	Feature Type	Feature Count	Target Variable	Instances	Official Train/Test Split	Data Source	Format	License	Access
Casting Product Quality Inspection Grayscale images of pump impeller castings with and without defects.	2020	Image	300x300 512x512	C (2)	7.348	✔️	Real	JPG	CC-BY-NC-ND 4.0	Link
GC10-DET Defect Location for Metal Surface Grayscale images of metal surfaces with various defect types and corresponding bounding box annotations.	2020	Image	Varying	C (10)	3.570	❌	Real	JPG, XML	?	Link
Mechanic Component Images Grayscale images of air conditioner pistons with various defect types.	2020	Image	86x90	C (3)	285	❌	Real	PNG	?	Link
Multi-Stage Continuous Flow Process Anonymized process data of a production line with quality measurements of part dimensions.	2020	Signal	116	-	14.088	❌	Real	CSV	?	Link
Plastic Extrusion Defects Process data of a plastic extrusion process.	2020	Signal	470	-	226.536	❌	Real	CSV	CC BY-NC-ND 4.0	Link
AITEX Grayscale images of textile fabrics with various defect types and corresponding segmentation masks.	2019	Image	4096x256	C (13)	245	❌	Real	PNG, Mask	?	Link
Deep PCB Grayscale images of circuit boards with various defect types and corresponding bounding box annotations.	2019	Image	640x640	C (7)	1.500	✔️	Real	JPG, Mask	only for research purpose	Link
Severstal Steel Defect Detection Grayscale images of steel surfaces with various defect types and corresponding segmentation polygons.	2019	Image	1600x256	C (5)	18.074	✔️ 🌐	Real	JPG, CSV	?	Link
Turning Dataset for Chatter Diagnosis Sensory data of a turning test rig and varying strengths of chatter.	2019	Signal	8	C (4)	> 10 Mio.	❌	Real	MAT	CC BY 4.0	Link
Magnetic Tile Defect Grayscale images of magnetic tile surfaces with various defect types and corresponding segmentation masks.	2018	Image	248x373	C (6)	1.344	❌	Real	JPG, PNG	?	Link
TIG Welding Grayscale images of a welding process with various defect types.	2018	Image	800x974	C (6)	33.254	✔️	Real	PNG, JSON	CC BY-SA 4.0	Link
Mining Process Process data of a mining process for impurity prediction in ore concentrate.	2017	Signal	24	R	737.454	❌	Real	CSV	CC0: Public Domain	Link
Bosch Production Line Performance Anonymized process data of production lines with and without defects.	2016	Signal	4264	C (2)	2.368.435	✔️ 🌐	Real	CSV	?	Link
WM811K Wafer Maps Defect matrices of semiconductor wafers with various defect types.	2014	2D Defect Matrix	Varying	C (9)	811.457	❌	Real	MAT	?	Link
NEU Surface Defect Database Grayscale images of metal surfaces with various defect types and corresponding bounding box annotations.	2013	Image	200x200	C (6)	1.800	❌	Real	BMP, XML	?	Link
Steel Plate Faults Geometric measurements of steel plates with various defect types.	2010	Signal	27	C (7)	1.941	❌	Real	CSV	?	Link
HCI Industrial Optical Inspection Synthetic grayscale images of textured surfaces with corresponding defect ellipses.	2007	Image	512x512	C (2)	16.100	✔️	Synthetic	PNG, Non-Standard	?	Link

Process Parameter Optimization

Name	Year	Feature Type	Feature Count	Instances	Official Train/Test Split	Data Source	Format	License	Access
Laser Welding Process parameter recordings for correlation with weld quality indicators such as weld depth and geometrical dimensions.	2020	Signal	13	361	❌	Real	XLS	CC BY 4.0	Link
3D Printer Process parameters of a 3D printer for correlation with print quality indicators such as roughness, tension and elongation.	2018	Signal	12	50	❌	Real	CSV	?	Link
Tool Path Generation Shape deviation measurements and corresponding simulated cutting conditions.	2018	Signal	9	4.968	❌	Real	CSV	CC BY 4.0	Link
Mercedes-Benz Greener Manufacturing Car feature configurations to be correlated with the required test time of the configurations.	2017	Signal	378	8.420	✔️ 🌐	Real	CSV	?	Link
SECOM Semiconductor process measurements and corresponding yields for determination of key factors to yield.	2008	Signal	591	1.567	❌	Real	Non-Standard	?	Link