Chemical engineers frequently work with data from customers, processes, sensors, research activities, and physical property databases. On this page, you will find ChE-relevant data sets that are not part of competitions. Sometimes these data sets require cleaning; if you only want ready-to-analyze data sets, look here for data science challenges and contests that are of interest to chemical engineers.
Companies are obviously reluctant to publicly share real, raw data from their processes, because they want to maintain a competitive edge. Some companies work with individual researchers, but they normalize data to remove identifying information before sharing or publishing. Plus, it's expensive to induce a wide range of failure modes in real processes. Because of issues with data secrecy, access and variety, academics frequently rely on synthetic process data. If you'd like to simulate your own process, try OpenModelica.
This Tennessee Eastman Process simulation features a typical chemical process with reactor, condenser, compressor, separator, and stripper.
Chen (2019) DataReith (2017) Data
Ricker (1996) Data
Matlab Code (UW)
Matlab Code (ASU)
F77 Code
F77 Python Wrapper
With 591 unlabeled inputs for 1567 semiconductor wafers, this dataset has a go/no-go (1463/104) outcome with 42,000 pieces of missing data (4.5%).
Data: Text FilesDescription
McCann (2008)
Munirathinam (2016)
Salem (2018)
Github: Ditsworth
Github: Nakshatra
Chugh (2018)
Find Other Work
After a chemical vapor deposition process, thickness is measured at nine locations on 184 wafers.
Data: Excel FileDescription
Nineteen process variables are measured for a water-air system in a multiphase flow facility in the Process System Engineering Laboratory of Cranfield University.
Data: Excel & Text FilesArticle: Journal Site
Data for a water treatment plant, including 22 process inputs, 7 output measurements & 9 measurements of performance.
Data: .dat FileDescription
Predict net hourly electrical energy output for a combined cycle power plant, using exhaust vacuum and 3 ambient variables (T, P, RH).
Data: Excel FileDescription
To compare strategies for control, researchers use a benchmark simulation of a wastewater treatment facility.
Data: BSM1 ModelData: BSM2 Model
Description
Find Other Work
This 16-feature numerical simulation emulates a gas turbine propulsion plant on a naval vessel.
Data: Text FileDescription
ENGIE is making data available from it's La Haute Borne wind farm, whose four turbines have provided "electricity to the equivalent of 7,300 people since 2009."
DataDescription
Equipment manufacturers are more likely to share data -- and researchers can afford to purchase equipment and collect their own data -- for individual unit operations, such as compressors, pumps, separators, power sources, and more. Because the effects are local and economic & environmental impacts are greatly reduced, users have an opportunity to test a range of operating conditions, implement design changes, or induce a wide range of faults.
Dowdy, Kawakita, Lange & Simmons (2018) performed a meta-analysis of 134 MFC experiments that use waste streams.
Data: Excel FileArticle: Journal Site
Article: Author Site
Evaluating an absorber in a CO2 capture process, Kachko et al (2015) built models using 6 variables (conductivity, pH, density, speed of sound, refractive index, NIR spectra).
Data: Excel FileArticle: Journal Site
Data from an ammeter and oscilloscope features are collected for a motor drive, while varying speed, load moment and load force for 11 states.
Data: Text FileDescription
Seventeen sensors are measured at different sampling rates (1, 10, 100Hz) for an hydraulic test rig.
Data: Text FilesDescription
Related Work
Data for individual sensors are often related to (1) long-term performance (drift) and reliability (failure modes) testing in the presence of a constant/controlled stimulus or (2) transient performance (response time) when sensors are presented with a step change in stimulus. For non-specific chemical sensors (or for chemical sensors with significant interfering substances), data sets often include results for a range of pure chemicals, mixtures, and flow conditions.
Data from 72 metal-oxide gas sensors that were deployed at six locations in a wind tunnel and exposed to 10 single-species chemicals.
Data FilesDescription
Article: Journal Site
Data over 36 months from 16 metal-oxide sensors presented with six different chemicals, each at six concentrations.
Data FilesDescription
Article 1: Journal Site
Article 2: Journal Site
Data from eight metal-oxide sensors presented with two gas sources that mix under turbulent conditions inside a wind tunnel.
Data FilesDescription
Article: Journal Site
Data from 16 metal-oxide sensors -- four unique sensor types -- exposed to gas mixtures at varying concentrations.
Data FilesDescription
Article: Journal Site
Simulating biological respiration, an external mechanical ventilator delivered chemical mixtures in the presence of 16 metal-oxide gas sensors.
Data FilesDescription
Article: Journal Site
Five Quartz Crystal Microbalance (QCM) gas sensors were exposed to 1-octanol, 1-propanol, 2-butanol, 2-propanol and 1-isobutanol.
Data: 5 Excel FilesDescription
To estimate molecular properties, we often use computationally expensive methods in quantum chemistry, but data science algorithms can reduce the time/expense of these calculations. This section focuses on microscopic molecular properties -- like dipole moments, polarizability, etc. -- while the next section centers on macroscopic molecular properties, like partition coefficients and biodegradability.
Measured physical properties are often related to fundamental molecular properties -- and sometimes to other measured attributes that are more readily available. This section focuses on macroscopic properties, such as critical temperatures, toxicity, and solubility.
Using nine molecular properties for a set of 2874 measured solubilities, Delaney (2004) performed linear regression to estimate aqueous solubility.
Data: Text FileArticle: Journal Site
Using molecular descriptors along with biodegradation experimental values of 1055 chemicals, Mansouri et al (2013) developed a QSAR model for biodegradability.
Data: Excel FileDescription
Article: Journal Site
Backed by data from 1200 organic photovoltaics, Nagasawa et al (2018) used data-driven molecular design to improve power conversion efficiencies.
Data: Text FileArticle: Journal Site
Using up to 81 features for 21,263 different superconductors, estimate critical temperature.
Data: Excel FilesDescription
Original Source
Multiple databases have been curated for atomic and molecular properties, including a combination of microscopic and macroscopic properties. This section features a handful of relevant data sources, which can be used in the two sections above.
Many organizations curate and host data sets, some of which have been highlighted above. These repositories have a wide range of data and are always expanding their breadth and depth. It might be worth taking a closer look at these sites, especially for biomolecular systems, medical, environmental, and satellite data, where are underrepresnted in previous sections on this page.
Social sciences account for 40+% of this 100,000-item collection, but the repository has 5,500+ datasets for chem, eng, env & med.
Search Dataverse
Since hosting its first data set in 1987, the UC-Irvine Machine Learning Repository has expanded to include ~500 data sets.
Search UCI ML Data
Datasets up to 2TB and in various formats can be stored and retrieved in this data repository. Relatively new and expanding quickly.
Search DataPort
Although it doesn't host data sets, the Center of Data Innovation highlights and links to a different data set each week.
Recent Highlights
More than 22,000 data sets, many of which are tagged for relevant areas of science.
Tag: ChemistryTag: Biology
Tag: Environment
More than 200 datasets, including PHM challenges, other contests, and more. Data mostly posted earlier than 2015.
Search DASHlink
The Amazon Web Services (AWS) Open Data Registry lets people share data, with usage examples listed in search results.
Tag: BioTag: Sustainability
Since Oct 2015, Jeremy Singer-Vine has been compiling data sources from his weekly "Data Is Plural" newsletter into a handy spreadsheet.
Google Doc
Google Dataset Search was publicly released on Jan 23, 2020, providing access to more than 25 million publicly available datasets.
SearchMail List Sub
FAQ