As one of the basic equipment for rail transit operation and maintenance in China, the online measurement system of train wheelset mainly measures the running state of train running parts, including wheelset profile, tread abrasion and axle temperature.The online measurement system of wheelset profile had been paid more and more attention because of its technical difficulty and its close relationship with wheel-rail dynamics. This paper analyzed the method of traceability calibration and the key problem of ensuring the long-term stability of the online measurement system of wheelset profile from an engineering example, introduced the key steps of parameter tracking and calibration based on neural network, and expounded the error compensation method and application effects of radial runout and axial displacement in the online measurement system of wheelset profile.