Test methodology for electromechanical actuators

Electromechanical actuators are highly complex non-linear devices that cannot be accurately modeled using only analytical formulations derived from first principles. When the application demands high model accuracy with a wide parametric range (and criteria) plus the need to take manufacturing/assembly variations associated with the asbuilt actuator into account, an empirical model based on extensive testing across the entire operating domain is the recommended approach. Since testing is an expensive, time consuming and laborious process, it is the aim of this research to determine efficient test methodologies (experimental designs) that would obtain the maximum information about actuator performance by means of a minimal number of tests. Current test standards are primarily designed to arrive at the actuator specifications by carrying out tests at either a single or a very limited set of test points. The results thus obtained are typically not valid across the entire operating domain of the actuator. Also these tests are performed for a very small set (one or two) of criteria. Furthermore most of this testing is conducted in terms of just one (occasionally two) control variables. As a result the full capability of the actuator is poorly represented. The research presented here addresses these limitations. To achieve the objective, the steps followed in this research are -- a) define a set of actuator performance criteria for testing, b) construct a test bed for actuator testing, c) develop a framework for testing actuators, d) conduct tests by applying principles from Design of Experiments, e) apply statistical techniques to identify empirical models and develop efficient experimental designs, and f) graphically present the actual capabilities of the actuator using performance maps. A commercially available permanent magnet synchronous motor-geartrain combination was chosen as the test actuator. This actuator has a nominal/peak rating of 43/86 lb-ft torque and 30/100 RPM speed. The criteria considered for characterizing the actuator’s operational capability includes noise, vibration, efficiency, current consumption, torque ripple, velocity ripple, backdriveability, and temperature. Control variables affecting the performance criteria were identified. Measurement of performance over the entire operating range of actuator requires that the actuator be operated at specific levels of these control variables and the concerned performance criteria be measured. Therefore to perform these actuator tests, a modular test bed was constructed. The test bed consists of an actuator loading mechanism (in the form of a magnetic particle brake or a geartrain-motor combination), an array of sensors, amplifiers, a signal conditioning unit, data acquisition modules, motion controller, and transformer. The measured sensor data is filtered through the signal conditioning unit (to remove noise) and digitized using the data acquisition modules. Statistical techniques were employed to process the sensor data and for each criterion, an empirical model relating the criteria to its control variables was determined. Model adequacy checks were carried out to ensure that the model did not violate important statistical assumptions and that it adequately represented the relationship between the input control variables and the output response (performance criteria). These models were used to generate performance maps for each criteria. Based on a predetermined set of run sizes, for each empirical model, alternate experimental designs were determined. Efficient experiment designs were identified by metrics such as -- Gefficiency, maximum prediction variance and average prediction variance. Besides the obvious advantage of arriving at complete and accurate performance profiles for the actuator undergoing tests (with minimal testing), the methodology could be applied to other actuators of a similar family. We might consider the methodology to be a subset of the general concept of metrology; i.e., the determination of as-built parameters vs. as designed parameters. Simplification techniques were applied to these models to remove unwanted model terms.