T.J. Crumpton
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
1 records found
1
Wind tunnel testing is a fundamental part in wind energy control validation. Despite that, facilities around the globe used for wind energy control experiments offer limited capability in generating spatially heterogeneous and non-stationary inflow conditions in a controlled and repeatable manner. The WindShaper facility in the WindAI Lab at the Delft Center for Systems and Control (DCSC) provides independent control of a 729 fan-array and therefore offers the potential for spatial and temporal flow control. This thesis investigates to what extent the WindShaper facility is able to generate prescribed heterogeneous and non-stationary reference profiles.
The high-dimensional actuator space was first reduced to three spatial input coefficients, from which a static input-output model was identified. The static map captured the general input output behaviour well but was not accurate enough for open-loop tracking. Fixed-input experiments showed that both stationary and non-stationary flow remained approximately temporally stable over a duration of 1800 s, suggesting that within-run drift is unlikely to be a dominant source of the observed tracking errors. Fixed-model feedback experiments showed that the identified static map contained sufficient sensitivity information for accurate reference tracking, while periodic tracking experiments showed the relevance of temporal dynamics.
To account for sensitivity to model uncertainty, the static input-output model was updated online through several adaptive least-squares approaches and combined with a regularized inverse control law. The recursive least-squares approach showed the strongest tracking improvement. Under nominal conditions, the recursive least-squares algorithm saw a tracking improvement of 33.2 % relative to the open-loop controller and of 57.5 % under enforced model mismatch conditions. Under enforced model mismatch conditions, ten repeated recursive least-squares experiments were found to be statistically consistent with the repeatability limit estimated from the data. Under stricter and more relevant repeatability conditions repeatability coverage was found to depend on operating conditions and no universal repeatability was observed.
The results show that the WindShaper facility is able to generate repeatable prescribed mean velocity profiles under the investigated operating conditions and that online model adaptation significantly improved robustness to model uncertainty. The achievable run-to-run repeatability depends on the operating conditions and the specific criterion used to assess it. ...
The high-dimensional actuator space was first reduced to three spatial input coefficients, from which a static input-output model was identified. The static map captured the general input output behaviour well but was not accurate enough for open-loop tracking. Fixed-input experiments showed that both stationary and non-stationary flow remained approximately temporally stable over a duration of 1800 s, suggesting that within-run drift is unlikely to be a dominant source of the observed tracking errors. Fixed-model feedback experiments showed that the identified static map contained sufficient sensitivity information for accurate reference tracking, while periodic tracking experiments showed the relevance of temporal dynamics.
To account for sensitivity to model uncertainty, the static input-output model was updated online through several adaptive least-squares approaches and combined with a regularized inverse control law. The recursive least-squares approach showed the strongest tracking improvement. Under nominal conditions, the recursive least-squares algorithm saw a tracking improvement of 33.2 % relative to the open-loop controller and of 57.5 % under enforced model mismatch conditions. Under enforced model mismatch conditions, ten repeated recursive least-squares experiments were found to be statistically consistent with the repeatability limit estimated from the data. Under stricter and more relevant repeatability conditions repeatability coverage was found to depend on operating conditions and no universal repeatability was observed.
The results show that the WindShaper facility is able to generate repeatable prescribed mean velocity profiles under the investigated operating conditions and that online model adaptation significantly improved robustness to model uncertainty. The achievable run-to-run repeatability depends on the operating conditions and the specific criterion used to assess it. ...
Wind tunnel testing is a fundamental part in wind energy control validation. Despite that, facilities around the globe used for wind energy control experiments offer limited capability in generating spatially heterogeneous and non-stationary inflow conditions in a controlled and repeatable manner. The WindShaper facility in the WindAI Lab at the Delft Center for Systems and Control (DCSC) provides independent control of a 729 fan-array and therefore offers the potential for spatial and temporal flow control. This thesis investigates to what extent the WindShaper facility is able to generate prescribed heterogeneous and non-stationary reference profiles.
The high-dimensional actuator space was first reduced to three spatial input coefficients, from which a static input-output model was identified. The static map captured the general input output behaviour well but was not accurate enough for open-loop tracking. Fixed-input experiments showed that both stationary and non-stationary flow remained approximately temporally stable over a duration of 1800 s, suggesting that within-run drift is unlikely to be a dominant source of the observed tracking errors. Fixed-model feedback experiments showed that the identified static map contained sufficient sensitivity information for accurate reference tracking, while periodic tracking experiments showed the relevance of temporal dynamics.
To account for sensitivity to model uncertainty, the static input-output model was updated online through several adaptive least-squares approaches and combined with a regularized inverse control law. The recursive least-squares approach showed the strongest tracking improvement. Under nominal conditions, the recursive least-squares algorithm saw a tracking improvement of 33.2 % relative to the open-loop controller and of 57.5 % under enforced model mismatch conditions. Under enforced model mismatch conditions, ten repeated recursive least-squares experiments were found to be statistically consistent with the repeatability limit estimated from the data. Under stricter and more relevant repeatability conditions repeatability coverage was found to depend on operating conditions and no universal repeatability was observed.
The results show that the WindShaper facility is able to generate repeatable prescribed mean velocity profiles under the investigated operating conditions and that online model adaptation significantly improved robustness to model uncertainty. The achievable run-to-run repeatability depends on the operating conditions and the specific criterion used to assess it.
The high-dimensional actuator space was first reduced to three spatial input coefficients, from which a static input-output model was identified. The static map captured the general input output behaviour well but was not accurate enough for open-loop tracking. Fixed-input experiments showed that both stationary and non-stationary flow remained approximately temporally stable over a duration of 1800 s, suggesting that within-run drift is unlikely to be a dominant source of the observed tracking errors. Fixed-model feedback experiments showed that the identified static map contained sufficient sensitivity information for accurate reference tracking, while periodic tracking experiments showed the relevance of temporal dynamics.
To account for sensitivity to model uncertainty, the static input-output model was updated online through several adaptive least-squares approaches and combined with a regularized inverse control law. The recursive least-squares approach showed the strongest tracking improvement. Under nominal conditions, the recursive least-squares algorithm saw a tracking improvement of 33.2 % relative to the open-loop controller and of 57.5 % under enforced model mismatch conditions. Under enforced model mismatch conditions, ten repeated recursive least-squares experiments were found to be statistically consistent with the repeatability limit estimated from the data. Under stricter and more relevant repeatability conditions repeatability coverage was found to depend on operating conditions and no universal repeatability was observed.
The results show that the WindShaper facility is able to generate repeatable prescribed mean velocity profiles under the investigated operating conditions and that online model adaptation significantly improved robustness to model uncertainty. The achievable run-to-run repeatability depends on the operating conditions and the specific criterion used to assess it.