Parameter identification with FB L_MC4P_DynamicParIdent

FB L_MC4P_DynamicParIdent can be used to determine the dynamic parameters of an axes group that are required for a torque feedforward control defined in FB L_MC4P_GroupTorquePreControl . This involves assigning an internally calculated identification trajectory (path curve) to the axes group and identifying the parameters during this motion.

In the course of this, not all mechanical variables can be identified independently of each other. For example, this issue affects the end effector and tool mass, which can only be estimated as a total amount when the tool is mounted. That is why parameters called "minimal parameters" are estimated within the identification FB.

Preconditions for identification
  • Robots, motors and gearboxes must be at operating temperature.
    This is necessary because the friction parameters to be identified are heavily dependent on the gearbox temperature.
  • The controllers (position and speed controllers) must be aligned.
    Otherwise, the behaviour of the robot will be too imprecise.
  • For good identification, the stimulation trajectory must be configured to allow the largest possible workspace for the robot.
  • Positions, speeds and accelerations must be in the size of productive operation.
  • The axis torque must be able to reach at least 50% of the rated drive torque during identification.
  • The following error that occurs without torque feedforward control must not exceed 400% of the required following error.
  • The torque must not be too noisy (<10% of M Rated ).
  • The torque must not exhibit strong vibrations (<10% of M Rated ).
  • The setpoint filter of the drives (FB L_MC1P_AverageFilterSetValues ) should be switched off or, if absolutely necessary, configured with the minimum possible filter depth.
  • The measurement noise of the actual velocity and the actual acceleration should be minimised.
    Adapt for the identification process or the actual value filter time available in the controller.
  • The override of the axes group must be at least 0.8 (= 80%).
    If this is not the case, the identification and verification trajectories will be travelled at reduced velocity and acceleration (e.g. for test purposes). However, this means that system stimulation is likely to be insufficient to estimate the parameters reliably.

Example for excessive vibration of the torque profile; no identification is possible with this signal.

Example for a good torque profile; identification is possible with this signal.

Run identification

If the above conditions are fulfilled, identification of the dynamic parameters can be started with FB L_MC4P_DynamicParIdent .

Verifying identification results

The following options are available for verifying the results:

  • Checking the results using a quality criterion
    FB L_MC4P_DynamicParIdent features the lrConvergence output, via which the identification convergence quality is output. Good identification should deliver values of > 75%.
    Function: The FB travels a verification path after identification. In the course of this, the expected torque is determined using the identified minimal parameters and compared with the actually occurring motor torque. The lower the deviation, the better the results.
  • Checking the results with a rough estimate:
    The minimal parameters of the models are estimated using known and estimated values for the system parameters (for details of the equations used for the calculation, see the respective parameter descriptions). This calculation essentially checks the magnitude of the system parameters and shows whether the identified parameters are in the range of the expected system parameters. However, it is not possible to convert the identified minimal system parameters back into physical parameters (masses, etc.).
Note!

Certain system parameters may only have a very minor influence on the motor torque compared to other variables. An example of this is a drive system with strong friction which only travels at a low rate of acceleration. In this case, the mass inertia of the axis barely has any effect on the motor torques and cannot be estimated reliably.

If such effects occur, the identified minimal parameters can deviate from the theoretically calculated comparison values.

  • Checking the results with a torque comparison:
    The torque profile calculated with function block L_MC4P_GroupTorquePreControl is compared with the torque profile that actually occurs when the robot performs a desired motion. For this purpose, the override of the function block can be set to zero so that no motor torque can be pilot-controlled. The profiles can be compared using a trace of the torque profiles.
    Checking results with a torque comparison
    The following image shows a very good correlation between the calculated torque (red) and the actual torque (blue). This means that the results of the identification were correct and that the torque feedforward control can be used.

Good correlation between the calculated torque (red) and the actual torque (blue).

    The following image shows an inadequate correlation between the calculated torque (red) and the actual torque (blue). This means that the results of the identification were incorrect and that the torque feedforward control cannot be used because it would make the reference behaviour of the system worse.

Inadequate correlation between the calculated torque (red) and the actual torque (blue).