Background
In most industrial robotics systems deployed in a production setting, there are environmental considerations that affect the performance of a robot during operation. Temperature fluctuation, vibration, sensor configuration, and variable lighting conditions are all factors that can produce undesirable—and often unacceptable—variance in the manufacturing process that the robotic system is attempting to perform. This process variance often requires cost adjustments and late design changes where engineers spend hours identifying the sources of process variance and eliminating them.
Approach
In this project, we investigated a method for modeling process variance and embodied it in a software framework to reduce the potential for undesirable variance. Using this model, we computed the gradient of the input factors that control the sources of process variance; this relative gradient allowed engineering staff to better understand which sources have the greatest impact. We also developed an optimization model to identify ideal values for the factors that control process variance and to minimize the financial cost of implementing those changes.
Accomplishments
We created a modular software framework for modeling process variance and optimizing the input parameters to minimize process variance and financial cost. We achieved close estimates of process variance using a model of a real system wherein process variance was measured with a metrology system. In this case, the primary sources of process error were vibration in two axes of a 3D sensor, the accuracy of the sensor, and the depth resolution of the sensor as a factor of its standoff distance. Despite challenges collecting the validation data and modeling complex process variance sources, the results of this project provide value as a preliminary check when designing a new system with known sources of process variance or when debugging issues with an existing system. These results also provide insight from the gradients about which sources of error are likely to have the greatest impact on reducing process variance and cost. In our test case, the vibration magnitude accounted for the largest process error gradient with a value of 2.1; the sensor accuracy gradient had a value of 1.0; and the sensor standoff distance had a gradient value of 0.002. From these results, we concluded that pursuing improvements to the stability of the sensor mount would produce the greatest reduction in process variance. Similarly, the variance and cost optimization framework can help identify ideal values for controllable parameters when choosing sensors or equipment for a new system design or modification. In our test case, we found that reducing vibration by 3 orders of magnitude and using a slightly less accurate sensor at a shorter standoff distance could improve process variance to an acceptable level while staying within the cost constraints of the application.