Klyachkin Vladimir Nikolaevich, doctor of technical sciences, professor, sub-department of applied mathematics and informatics, Ulyanovsk State Technical University (32 Severny Venec street, Ulyanovsk, Russia), E-mail: firstname.lastname@example.org
Zhukov Dmitry Anatolyevich, database specialist, Ulyanovsk branch of the design bureau of PJSC Tupolev (1 O. K. Antonova avenue, Ulyanovsk, Russia), E-mail: email@example.com
Background. To ensure the reliability of the technical object, its diagnostics is carried out in operating conditions according to the results of monitoring the performance indicators of this object. The relevance of the task is due to the constantly growing requirements for the safety and reliability of technology. When diagnosing the operability of a technical object, it is required to evaluate its condition according to specified performance indicators. In this case, multidimensional classification methods can be used, both standard statistical and special machine learning methods. Significant features of the problem under consideration are, firstly, the imbalance of the training sample: information on the performance indicators for inoperative states of the object is much less than for healthy ones, and secondly, a relatively small sample size (usually hundreds of observations, unlike “BigData” «Tens and hundreds of thousands in the usual problems of machine learning). The aim of the study is to develop a software package that, automatically, analyzing the initial data on the results of the previous operation, would give an opinion on the operability of the facility and would predict its condition according to specified performance indicators.
Materials and methods. To solve the diagnostic problem, various machine learning methods can be used: a naive Bayesian classifier, a support vector method, compositional methods (bagging and boosting) and others. To increase the accuracy of diagnostics, aggregated classification methods can be applied that use a combination of basic models built on a training set. When forming a unified decision on the operability of an object, you can use three options for aggregating results: by average, by median, and also by using the voting procedure. In addition to aggregation, the quality of diagnostics also depends on the selected volume of the control sample and the method of selecting significant indicators. The share of the control sample in the total volume of the source data (or the number of blocks of sample splitting during crossvalidation) has an ambiguous effect on the quality of classification: for each specific technical object, it is necessary to evaluate this factor individually. Another factor that significantly affects the quality of diagnostics is the significance of the influence of the considered indicators of functioning on the state of the object. Two approaches to assessing the significance of indicators are considered.
Results and conclusions. An algorithm and software package for diagnosing the technical condition of an object using aggregated classifiers has been developed, including: dividing the source data into control and training samples, selecting significant features, building basic and aggregated classifiers, searching for the best method according to the F-criterion, as well as the ability to predict state of the technical object. All these functions are calculated in an automated mode, thus, to carry out diagnostics of a technical object, it is enough to download data on its previous states and the software package will select the optimal parameters to obtain the most accurate result. The hot water supply system in the city of Ulyanovsk was used as the object of study: control was carried out according to data taken from water supply meters. By carrying out statistical tests, the effectiveness of the developed models and algorithms is shown, while the value of the Fcriterion in the studied samples through the use of aggregation, the choice of control volume and the selection of significant indicators increased to 15 % relative to the basic methods.
1. Mishanov R. O. Nadezhnost' i kachestvo slozhnykh system [Reliability and quality of complex systems]. 2019, no. 4 (28), pp. 112–117. [In Russian]
2. Severtsev N. A., Betskov A. V., Dar'ina A. N. Nadezhnost' i kachestvo slozhnykh system [Reliability and quality of complex systems]. 2019, no. 2 (26), pp. 19–26. [In Russian]
3. Yurkov N. K. Nadezhnost' i kachestvo slozhnykh system [Reliability and quality of complex systems]. 2014, no. 1 (5), pp. 18–24. [In Russian]
4. Birger I. A. Tekhnicheskaya diagnostika [Technical diagnostics]. Moscow: Mashinostroenie, 1978, 240 p. [In Russian]
5. Klyachkin V. N., Krasheninnikov V. R., Kuvayskova Yu. E. Prognozirovanie i diagnostika stabil'nosti funktsionirovaniya tekhnicheskikh ob"ektov [Forecasting and diagnostics of the stability of the functioning of technical objects]. Moscow: Rusayns, 2020, 200 p. [In Russian]
6. Klyachkin V. N., Karpunina I. N. International Journal of Innovative Technology and Exploring Engineering (IJITEE). 2019, vol. 8 (6S3), pp. 590–593.
7. Voronina V. V., Mikheev A. V., Yarushkina N. G., Svyatov K. V. Teoriya i praktika mashinnogo obucheniya: ucheb. posobie [Theory and practice of machine learning: a textbook]. Ulyanovsk: UlGTU, 2017, 290 p. [In Russian]
8. Neykov M., Liu Jun S., Cai Tianxi Journal of Machine Learning Research. 2016, no. 17 (70), pp. 1–32.
9. Wyner A. J., Olson M., Bleich J., Mease D. Journal of Machine Learning Research. 2017, no. 18 (48), pp. 1–33.
10. Klyachkin V. N., Kuvayskova Yu. E., Zhukov D. A. Radiotekhnika [Radio engineering]. 2018, no. 6, pp. 46–49. [In Russian]
11. Klyachkin V. N., Kuvayskova J. E., Zhukov D. A. International Multi-Conference on Industrial Engineering and Modern Technologies (FarEastCon – 2019). Vladivostok, 2019, vol. 22, pp. 244–260. DOI 10.15827/0236–235x.126.244–260.
12. Zhukov D. A. Sovremennye problemy proektirovaniya, proizvodstva i ekspluatatsii radiotekhnicheskikh sistem: sb. [Modern problems of design, production and operation of radio engineering systems: collection]. Ulyanovsk, 2019, pp. 214–216. [In Russian]
13. Davis J., Goadrich M. Proceedings of the 23rd international conference on Machine learning. Pittsburgh, 2006, pp. 233–240.
14. Zhukov D. A. Avtomatizatsiya protsessov upravleniya [Automation of management processes]. 2019, no. 3 (57), pp. 112–117. [In Russian]