A Neural Network Mathematical Model Design Method for Estimating Compressed Air Volume Flow through a Proportional Valve

The article describes an approach to neural network model design for simulating processes in shut-off and control pneumatic devices. This type of model can be used for a reasoned selection of components for multi-component pneumatic system configurations. As an example, the application of the proposed approach to the development of an artificial neural network to estimate the compressed air volume flow through a proportional valve is considered. The manufacturer’s catalog is used to obtain data samples. The structure of the proposed neural network model, data preprocessing for model configuration, and the selected learning algorithm are described. A computer program for compiling train and test data samples and the subsequent neural network training is developed. The results of measurements are simulated using additional, normally distributed noise with a standard deviation of 0.02. The results obtained using two mathematical models, the neural network model and the classical one, supplemented by empirical coefficients, are compared. The maximum deviation between the two models is less than 1.5 % of the maximum volume flow rate for a particular proportional valve model.

References

[1] Camozzi Automation. Pnevmaticheskaya apparatura. Bol’shoy katalog. Versiya 8.8. Available at: http://catalog.camozzi.ru/pdf/series_ap.pdf (accessed 11 February 2019).

[2] Compact Proportional Solenoid Valve. Series PVQ. Available at: http://stevenengineering.com/Tech_Support/PDFs/70PCPVQ.pdf (accessed 11 February 2019).

[3] IMI Norgren. VP40, 2/2 — Proportional flow control valve. Direct actuated poppet valve (stainless steel). Available at: http://cdn.norgren.com/pdf/en_6_6_024_VP40.pdf (accessed 11 February 2019).

[4] Proportion-Air, Inc. Aerospace solutions. Available at: https://proportionair.com/markets/aerospace/ (accessed 11 February 2019).

[5] Arzumanov Yu.L., Khalatov E.M., Chekmazov V.I., Chukanov K.P. Matematicheskie modeli sistem pnevmoavtomatiki [Mathematical models of pneumatic automation systems]. Moscow, Bauman Press, 2009. 296 p.

[6] Gradetskiy V.T., Dmitriev V.N. Osnovy pnevmoavtomatiki [Basics of pneumatic automation]. Moscow, Mashinostroenie publ., 1979. 360 p.

[7] Zelenov M.S., Atamasov N.V., Chernyshev A.V. On the issue of simulating dynamics of a pneumo-mechanical device. Herald of the Bauman Moscow State Technical University. Series Mechanical Engineering, 2018, no. 6, pp. 20–33 (in Russ.), doi: 10.18698/0236-3941-2018-6-20-33

[8] Grishelenok D.A., Kovel’ A.A. Application of mathematical planning of experiment method for neural network training. Journal of Instrument Engineering, 2011, vol. 54, no. 4, pp. 51–54 (in Russ.).

[9] Nikolenko S., Kadurin A., Arkhangel’skaya E. Glubokoe obuchenie. Pogruzhenie v mir neyronnykh setey [Deep learning. Immersion in the world of neural networks]. Sankt-Petersburg, Piter publ., 2018. 480 p.

[10] Burakov M.V. Neyronnye seti i neyrokontrollery [Neural networks and neurocontrollers]. Sankt-Petersburg, GUAP publ., 2013. 284 p.

[11] Hao Yu, Wilamowski B.M. Levenberg–Marquardt training. Industrial Electronics Handbook, Second Edition: Intelligent Systems, CRC Press, 2011, vol. 5, pp. 12.1–12.15.