<p>The technique of electrical impedance tomography (EIT) has been recognized as a promising method to design tactile sensors with continuous sensing capability over a large area. The mechanism of electrical impedance tomography allows reconstructing tactile information within the sensing area based on measurements made only at the boundary. However, spatial performance of EIT-based tactile sensors has demonstrated location dependency in previous reports, which severely affects correct interpretation of tactile stimuli.


The objective of this dataset is the fault diagnosis in diesel engines to assist the predictive maintenance, through the analysis of the variation of the pressure curves inside the cylinders and the torsional vibration response of the crankshaft. Hence a fault simulation model based on a zero-dimensional thermodynamic model was developed. The adopted feature vectors were chosen from the thermodynamic model and obtained from processing signals as pressure and temperature inside the cylinder, as well as, torsional vibration of the engine’s flywheel.




This dataset is used for i) analyzing the influence of process information on monitoring signals through signal processing methods; ii) training and testing models of tool monitoring and tool wear prediction especially for cutting conditions with large variations including cutting parameters, material and geometry of cutting tools, and workpiece materials, and also cutting conditions with continuous changes. This data set includes monitoring signals collected from machining process of sidewalls and closed pockets. The sidewall machining belongs to the cutting process with fixed cutting conditions; the closed pocket machining belongs to the cutting process of continuously varying cutting conditions for the reason that the tool path of closed pocket includes line, arc, full cutting and non-full cutting. Although cutting parameters are given fixed in the arc tool path area, the actual cutting parameters (such as feed, cutting width) are constantly changing due to the change of cutting geometry.


The datasets are related to the findings and results of our investigations of the minimal force thresholds perception in robotic surgical applications. The experimental setup included an indenter-based haptic device acting on the fingertip of a participant and a visual system displays grasping tasks by a surgical grasper. The experiments included the display of a set of presentations in three different modes, namely, visual-alone, haptic-alone, and bimodal (i.e., combined). Sixty participants took part in these experiments and were asked to distinguish between consecutive presentations.


This folder contains the onboard sensor measurements for the EksoGT robotic exoskeleton during experiments with three able-bodied individuals and three non-able-bodied individuals with a spinal cord injury. The data is divided into .mat files by the trial. All able-bodied subjects completed three repetitions of each commanded intent change (Speed Up, Slow Down, and No Change) for each trial set.


Please see ReadMe.pdf for details about the dataset.


Acoustic measurement data from Multilayer Ceramic Capacitors (MLCCs). Contains preprocessed data from intact and damaged MLCCs for damage detection (classification) purposes.


Contains acoustic measurement data from 180 multilayer ceramic capacitors (2220 case size, 22 uF, 24V), soldered onto two test circuit boards. The measurements were performed by placing a piezoelectric point contact sensor on top of each capacitor, and subjecting the MLCC to a voltage frequency sweep from 100 Hz to 2 MHz over a duration of 100 ms. The resulting acoustic waveforms have been denoised, bandpass filtered, and downsampled. Furthermore, instantaneous phase response was calculated for each MLCC.

The dataset contains measurements from both intact and mechanically damaged components for quality assurance purposes (classification task). The acoustic signature of each MLCC is represented by an eight-dimensional feature vectror in the file inputs.mat:

  1. Acoustic emission amplitude at the highest resonace peak
  2. Frequency of the highest resonance peak
  3. Amplitude of the second-highest resonance peak
  4. Frequency of the second-highest resonance peak
  5. Total phase shift during frequency sweep
  6. Median amplitude of 10 of the highest resonance peaks
  7. Median frequency of 10 of the highest resonance peaks
  8. Mean group delay ripple calculated from the phase response of each component

The labels (0=no damage; 1=damage) for each component are found in targets.mat. Note that the labelling process was done by cross-sectioning each component and inspecting the sample visually under a microscope. Therefore, the labels may not be completely accurate, as the signs of damage can be difficult to observe.



This dataset is used to develop an algorithm for evaluating machining quality. When machining a workpiece in a milling process, vibration signals can be recorded by a 3-axis accelerometer, which is attached on the spindle of a CNC milling machine. To evaluate machining quality, the vibration signals can be segmented and extracted the corresponding features, in the time, frequency, and time-frequency domains. After serving with the features, a model can be developed to estimate the machining quality, such as the roughness of a workpiece.


Domain: milling process

Signal source: accelerometer (ITRI)

Sampling rate: 2048 Hz

Target: quality estimation


  • Spindle Speed (rpm): 3000
  • Feed Rate (mm/min): 800
  • Workpiece: FDAC
  • Tool Diameter (mm): 6
  • No. of Flutes: 4

No. of file: 1

File format: csv

Description of fields in each file

  • Sample_No: sampling no, time interval 1 ms
  • Rawdata_X: vibration signals of axis X. 
  • Rawdata_Y: vibration signals of axis Y.
  • Rawdata_Z: vibration signals of axis Z.  

This dataset is used to develop an algorithm for automatic segmenting the collected signals. When machining a workpiece in a milling process, vibration signals can be recorded by a 3-axis accelerometer, which is attached on the spindle of a CNC milling machine. To segment the recorded signals, a moving window (0.5 sec) is applied to sample the vibration signals and manually labeled the corresponding modes, i.e. dry run or milling, of each window. To verify the algorithm, 3 types of operations are provided and recorded in csv format. 


Domain: milling process

Signal source: accelerometer 

Sampling rate: 2048 Hz 

Target: signal segmentation


  1. EXP_Drill: 
  • Spindle Speed (rpm): 2300
  • Feed Rate (mm/min): 190
  • Workpiece Material: SUS316
  • Ap (mm): 10
  • Ae (mm): 8.6
  • Tool Type: Drill
  • Tool Diameter (mm): 8.6
  • No. of Flutes: 2
  • No. of Files: 15


  • Spindle Speed (rpm): 6000
  • Feed Rate (mm/min): 1500
  • Ap (mm): 5
  • Ae (mm): 5
  • Workpiece Material: AL6061
  • Tool Type: Milling
  • Tool Diameter (mm): 10
  • No. of Flutes: 3
  • No. of Files: 2


  • Spindle Speed (rpm): 1600
  • Feed Rate (mm/min): 320
  • Workpiece Material: FDAC
  • Ap (mm): 1
  • Ae (mm): 5
  • Tool Type: Milling
  • Tool Diameter (mm): 10
  • No. of Flutes: 4
  • No. of Files: 8

Total files: 25

File format: csv

Description of fields in each file

  • Timetag: tagging sequence per 0.5 sec
  • Segmentation: 0: dry run, 1: milling
  • Rawdata_X_1 ~ Rawdata_X_1024: vibration signals of axis X recorded at sampling rate 2048 Hz. It means the data collected in 0.5 sec.  
  • Rawdata_Y_1 ~ Rawdata_Y_1024: vibration signals of axis Y.
  • Rawdata_Z_1 ~ Rawdata_Z_1024: vibration signals of axis Z.  



This dataset was collected from force, current, angle (magnetic rotary encoder), and inertial sensors of the NAO humanoid robot while walking on Vinyl, Gravel, Wood, Concrete, Artificial grass, and Asphalt without a slope and while walking on Vinyl, Gravel, and Wood with a slope of 2 degrees. In total, counting all different axes and components of each sensor, we monitored 27 parameters on-board of the robot.


When producing bolts in a cold forging process, the pressure signals are recorded per cycle of forming a bolt. The dataset is collected from experiments of different failure modes of a forming machine. Two experiments were recorded in csv format for providing four failure modes, including core broken, cavity block, insufficient lubrication, and material out-of-specification, as well as one normal mode. The two experiments were performed in the same machine with different cavities and cores, and saved in Experimental Data for Modeling and Testing.

  • Domain: cold bolt forging process
  • source: pressure sensor
  • Sampling rate: 1000 Hz
  • Target: failure classification
  • Failure modes:
    • 1. Normal: normal production.
    • 2. Cavity Block: a cavity was blocked with different heights of bumps.
    • 3. Core Broken: a core being broken into three levels
    • 4. Insufficient Lubrication: insufficient lubrication in two levels.
    • 5. Material out-of-specification: the materials being shorter or longer.
  • Total files: 2 
  • File format: csv
  • Description of fields in each file
    • TimeTag: experimental recording time of each producing cycle.
    • Label: faliure mode
    • Rawdata1-Rawdata386: pressure data collected by each bolt forming in a sampling rate 1000 Hz. 1-386 means it takes 0.386 sec to finish a bolt forming cycle, and their values stand for the raw data without standardization.