Computers & Chemical Engineering

The supplementary data includes the raw data for fitting the cell density vs. absorbance, and the experimental data of Plectonema growth profiles used in the article.

The dynamic process simulator K-Spice, has been used to simulate offshore high pressure produced water treatment. The initial conditions: feed temperature, pressure and oil to water ratio have been sampled with an actuator step change. Latin Hypercube Sampling has been used to produce 1000 samples per actuator. The system contains six actuators, of which all are valves, resulting in a total of 6000 samples. Each ".hst" file contains the time series from the system process varaibles.

The Excel-file contains the cost coeffients and other data not given in the manuscript.

The data was generated randomly, based on different combinations of possible actions in the dynamic simulator K-Spice from Kongsberg Digital. In case study 1, the aim was to increase +10 % of the oil production with respect to the initial condition value. In case study 2, the aim was to decrease -10 % of the gas production with respect to the initial condition value. The data show examples of possible correct and incorrect paths that a trainee could follow trying to solve the scenarios.

1. Description of supplementary material 2. Excel files containing mGBD algorithm (zipped) 3. Aspen Plus Version 8.8 files (zipped)

This data was generated using an advanced mathematical simulation of a 100,000 litre penicillin fermentation system referenced as IndPenSim. All details describing the simulation are available on the following website: www.industrialpenicillinsimulation.com. IndPenSim is the first simulation to include a realistic simulated Raman spectroscopy device for the purpose of developing, evaluating and implementation of advanced and innovative control solutions applicable to biotechnology facilities. This data set generated by IndPenSim represents the biggest data set available for advanced data analytics and contains 100 batches with all available process and Raman spectroscopy measurements (~2.5 GB). This data is highly suitable for the development of big data analytics, machine learning (ML) or artificial intelligence (AI) algorithms applicable to the biopharmaceutical industry. The 100 batches are controlled using different control strategies and different batch lengths representing a typical Biopharmaceutical manufacturing facility: Batches 1-30: Controlled by recipe driven approach Batches 31-60: Controlled by operators Batches 61:90: Controlled by an Advanced Process Control (APC) solution using the Raman spectroscopy Batches 91:100: Contain faults resulting in process deviations. Please reference: Goldrick S., Stefan, A., Lovett D., Montague G., Lennox B. (2015) The development of an industrial-scale fed-batch fermentation simulation Journal of Biotechnology, 193:70-82. and Goldrick S., Duran-Villalobos C., K. Jankauskas, Lovett D., Farid S. S, Lennox B., (2018) Modern day control challenges for industrial-scale fermentation processes. Computers and Chemical Engineering. Additionally help publicise this work on the following platforms: Twitter: @Stephen_Goldric Github: StephenGoldie LinkenIn: Stephen Goldrick - Post Doc @UCL Biochemical Socitey

NIR spectral data obtained for the development of calibration models and their validation. The data included is not pre-processed. This is the data obtained from the NIR instrument ("raw data").

Phase stability analysis, matlab codes and lingo code

Extended Tennessee Eastman Reference Data as presented in Reinartz et al. (2021). The data contains simulations of 28 process faults for 6 different operating modes. Each fault is simulated 500 times with different seeds for the random number generator. Additional simulations feature setpoint changes and mode transitions between the 6 operating modes. All simulations have a duration of 100 hours with a sampling rate of 3 minutes. Please refer to the Readme and publication for detailed information.

Extended Tennessee Eastman Reference Data as presented in Reinartz et al. (2021). The data contains simulations of 28 process faults for 6 different operating modes. Each fault is simulated 500 times with different seeds for the random number generator. Additional simulations feature setpoint changes and mode transitions between the 6 operating modes. All simulations have a duration of 100 hours with a sampling rate of 3 minutes. Please refer to the Readme and publication for detailed information.