Dark Mode
Prognostics and Health Management (PHM) Filter Dataset
Patient Health Records & Digital Health
Tags and Keywords
Trusted By
"No reviews yet"
Free
About
Machine learning applications in Prognostics and Health Management (PHM), specifically focusing on predicting the Remaining Useful Life (RUL) of filters. The data originates from a practical degradation process: the clogging of filters as they separate solid particles from gas. It was generated using a test bench that performs automated life testing of filter media by loading it with dust complying with ISO standard 12103-1. The primary challenge addressed by this dataset is the issue of varying data quality, which simulates real-world scenarios where sensors, their positioning, or peripherals differ across systems. The dataset has been intentionally manipulated to include varying levels of signal-to-noise ratios and biases across different run-to-failure trajectories, making it ideal for developing robust models that can handle such variations. The goal is to accurately predict the RUL of filters based on condition data, primarily the differential pressure, which indicates clogging.
Columns
- Differential Pressure: Represents the condition data of the filter. Failure occurs when this value exceeds 600 Pa. Measurements are intentionally obscured by noise and shifted by one of four bias levels.
- Flow Rate: The rate of gas flow through the filter system.
- Dust Feed: The amount of dust supplied to the filter over time, which is a key factor in the clogging process.
- Current Time: The timestamp for each data point in a run-to-failure cycle.
- Remaining Useful Life (RUL): The target variable for prediction, indicating the time remaining until filter failure. This is provided for each data point in the training set and for the last point of censored trajectories in the test set.
Distribution
The dataset is distributed as
.mat and .csv files. The training data consists of 35 run-to-failure cycles, while the test data contains 20 randomly right-censored run-to-failure trajectories. The primary data file, Data.mat, has a size of 839.27 kB. The dataset is expected to be updated quarterly.Usage
This dataset is ideal for developing and testing models for Remaining Useful Life (RUL) prognosis. The main task is to predict the RUL of the censored filter test cycles in the test data. It is particularly useful for building models that can handle varying data qualities, such as different levels of noise and bias, which is a common challenge in industrial applications. The data also allows for the integration of physical knowledge or models, supporting approaches like theory-guided data science or informed machine learning.
Coverage
The data was generated on a single filter test bench located in Göppingen, Germany, by the Hochschule Esslingen - University of Applied Sciences. The dataset does not have a specific geographic or demographic scope beyond the controlled experimental setup. The time range pertains to the duration of the run-to-failure cycles conducted on the test bench.
License
Attribution 4.0 International (CC BY 4.0)
Who Can Use It
- Data Scientists & Machine Learning Engineers: For developing and benchmarking predictive maintenance and RUL prediction models, especially those robust to data quality variations.
- Researchers in PHM: For studying degradation processes and exploring theory-guided data science approaches by combining data-driven models with physical knowledge of filter clogging.
- University Students: As a practical dataset for projects in time series analysis, regression, and predictive analytics.
Dataset Name Suggestions
- Filter Clogging RUL Prediction with Varying Data Quality
- Predictive Maintenance: Filter Degradation Dataset
- Sensor Data for Filter Remaining Useful Life Prognosis
- Industrial Filter Clogging Time Series Data
- Prognostics and Health Management (PHM) Filter Dataset
Attributes
Original Data Source: Prognostics and Health Management (PHM) Filter Dataset
QUALITY