Friction compensation and external force estimation for robotic systems using a fuzzy neural network approach

Wan, Jun (author), Zhou, ZiHao (author), Yun, Nuo (author), Zhang, Xiao Yong (author), Tang, Jinlong (author), Wang, Kehong (author)

To address inaccurate external force estimation caused by nonlinear friction in robotic systems, this paper proposes a friction compensation and external force estimation method based on an adaptive neuro-fuzzy inference system (ANFIS). The approach integrates Stribeck friction modeling with a Takagi–Sugeno fuzzy inference structure to identify joint friction parameters from measured data. Experimental results show that ANFIS yields lower identification errors and better generalization performance than baseline methods including fuzzy neural networks, particle swarm optimization, and least squares. The implemented feedforward compensation strategy achieves maximum torque errors of 0.263 Nm and 0.184 Nm for the two joints, lower than those obtained by the compared approaches. By incorporating the identified friction model into a generalized momentum observer with median and Butterworth filtering, the proposed method reduces the root mean square error and maximum absolute error by 18.3 % and 27.9 %, respectively, and achieves a coefficient of determination (R²) of 0.994. In collision detection tests, the method identifies impact events with reduced false alarm rates under the same experimental settings, supporting its applicability to high-precision force control in robotic applications.

• Language: English
• Keywords: Stribeck model, fuzzy neural networks, friction compensation, external force estimation
• Work type: Article
• Typology: 1.01 - Original Scientific Article
• Organization: FS - Faculty of Mechanical Engineering
• Publication status: Published
• Publication version: Version of Record
• Year: 2026
• Number of pages: Str. 40-51
• Numbering: Vol. 72, no. 1/2
• PID: 20.500.12556/RUL-181068
• UDC: 531.43:004.032.26:007.52
• ISSN on article: 0039-2480
• DOI: 10.5545/sv-jme.2025.1489
• COBISS.SI-ID: 272814339
• Publication date in RUL: 24.03.2026

Record is a part of a journal

• Title: Strojniški vestnik = Journal of mechanical engineering
• Shortened title: Stroj. vestn.
• Publisher: Univerza v Ljubljani, Fakulteta za strojništvo = University of Ljubljana, Faculty of Mechanical Engineering, Univerza v Mariboru, Fakulteta za strojništvo = University of Maribor, Faculty of Mechanical Engineering, Zveza strojnih inženirjev Slovenije = Association of Mechanical Engineers of Slovenia, Gospodarska zbornica Slovenije, Združenje kovinske industrije = Chamber of Commerce and Industry of Slovenia, Metal Processing Industry Association
• ISSN: 0039-2480
• COBISS.SI-ID: 762116

Licences

• License: CC BY 4.0, Creative Commons Attribution 4.0 International
• Link: http://creativecommons.org/licenses/by/4.0/
• Description: This is the standard Creative Commons license that gives others maximum freedom to do what they want with the work as long as they credit the author.

Secondary language

• Language: Slovenian
• Title: Kompenzacija trenja in ocena zunanjih sil v robotskih sistemih z uporabo mehkih nevronskih mrež

Abstract

Za odpravo nenatančne ocene zunanjih sil, ki je posledica nelinearnega trenja v robotskih sistemih, je predlagana metoda kompenzacije trenja in ocene zunanjih sil na osnovi prilagodljivega nevro-mehkega inferenčnega sistema (ANFIS). Pristop združuje Stribeckov model trenja s Takagi–Sugeno mehko-inferenčno strukturo za identifikacijo parametrov trenja v sklepih na podlagi izmerjenih podatkov. Eksperimentalni rezultati kažejo, da ANFIS dosega manjše identifikacijske napake in boljšo sposobnost posploševanja v primerjavi z referenčnimi metodami, kot so mehke nevronske mreže, optimizacija z rojem delcev in metoda najmanjših kvadratov. Implementirana kompenzacijska strategija vnaprejšnjega (feedforward) krmiljenja doseže največje napake navora 0,263 Nm in 0,184 Nm za oba sklepa, kar je manj kot pri primerjanih pristopih. Z vključitvijo identificiranega modela trenja v posplošeni opazovalnik momenta s filtrom na osnovi mediane in Butterworthovim filtriranjem predlagana metoda zmanjša srednjo kvadratno napako (RMSE) in največjo absolutno napako za 18,3 % oziroma 27,9 % ter doseže koeficient determinacije (R²) 0,994. V preizkusih zaznavanja trkov metoda zazna udarne dogodke z nižjo stopnjo lažnih alarmov pri enakih eksperimentalnih pogojih, kar potrjuje njeno uporabnost za visoko natančno krmiljenje sil v robotskih aplikacijah.

• Keywords: Stribeckov model, mehke nevronske mreže, kompenzacija trenja, ocena zunanjih sil

Projects

• Funder: Other - Other funder or multiple funders
• Funding programme: National Natural Science Foundation of China
• Project number: 52305381

• Funder: Other - Other funder or multiple funders
• Funding programme: Fundamental Research Funds for the Central Universities
• Project number: 30924010937