fb_epid Development History

Table of contents

• Testing Progression
  • Initial testing (June 2004)
  • First real feedback test (July 2004)
  • Second test round (October 2004)
• Key Tuning Findings
  • PID parameters
  • Bumpless start
  • Update rate

This page summarizes key findings from the initial development and testing of the fb_epid support in 2004, and the subsequent revision into its current form.

Testing Progression

Initial testing (June 2004)

The initial testing used a swait record to simulate a crystal reflection with a Lorentzian peak shape. An EPID record was configured for closed-loop feedback in either PID or Max/Min mode.

This testing revealed that the bare EPID record needed a wrapping database to be useful in practice. The following features were identified as necessary:

• Automatic on/off control based on external conditions
• User-selectable input and output PVs at runtime
• Separate tuning parameters for PID vs Max/Min modes
• Bumpless start (preventing positioner jumps when enabling feedback)
• Output limits loaded from the positioner

First real feedback test (July 2004)

The first real feedback test was conducted on APS beamline 33ID, maintaining the Oxford XBPM vertical position by adjusting the DCM omega2 piezo. Key observations:

• Response locked to a new position 1 mm away in approximately 10 seconds
• Position noise was observed when the PZT was moved; otherwise quieter than standard feedback

Second test round (October 2004)

The second round of testing confirmed stable operation with active XBPM-based feedback during sscan step-scans. The database was then revised to the present fb_epid form.

Key Tuning Findings

The following practical tuning guidance emerged from the initial testing:

PID parameters

• KP sign – KP must be negative when positive KP drives the positioner away from the setpoint. This is geometry-dependent.
• KI value – KI=400 gave mild overshoot. KI=1000 caused approximately 10 cycles of damped ringing before settling. KD=0 worked well in the tested configuration.

Bumpless start

Before enabling feedback (FBON), the I term must be preloaded with the current positioner value. The output limits (DRVL, DRVH) must also be loaded from the positioner. Without this, the positioner will jump to an arbitrary position when feedback is first enabled.

The current fb_epid database handles this automatically through the resume and outpv records (the “bumpless start” feature).

Update rate

The input signal must update 3 to 10 times faster than the EPID recalculation interval, or instability will occur.