A robotics pilot has to survive the whole shift
Cycle time can conceal the cost of stops, rejected parts and expert intervention. A deployment decision needs evidence from a complete operating period.
Consider an inspection cell that handles familiar parts quickly but stops when a reflective surface confuses its camera. The robot's best cycle time excludes the operator's walk to the cell, the inspection of the failed attempt and the restart. Faster arm movement may do less for output than better lighting or a different way to present the part.
A demonstration shows what the robot can do under test conditions. A customer needs to know how that translates into production. The pilot must reveal what happens across a full operating period, including variation, interruptions and recovery.
The part does not arrive alone
Define the operating envelope before the trial: part families, tolerances, surface finishes, presentation and environmental conditions. Include the interfaces with people and neighbouring equipment. A successful pick may still damage a critical surface or leave the part in an orientation the next machine cannot use.
NIST's perception research addresses ambient light, geometric features and flexible parts because a sensor's performance depends on the physical scene. An image classification result therefore cannot stand in for a manipulation trial. Build the test around variation in the customer's production, including supplier batches, fixture wear and permitted changes in illumination.
Keep conditions used to tune the system separate from those used to assess it. Retain failed attempts in the record. Simulation can help examine layouts and unusual conditions, but friction, contact and camera behaviour need physical checks where they affect the result. NIST's agility programme combines virtual and physical environments in its measurement work.
Count the time between successful cycles
Agree acceptance criteria with the production supervisor, maintenance team and quality owner. Their requirements can differ from those of an innovation sponsor. A cell that demonstrates a new capability may still require too much supervision to join the line.
Record completed acceptable units against scheduled operating time. Separate planned changeovers from unexpected stops, and capture the duration and cause of each intervention. Report unusually slow cycles and long interruptions alongside the average. A line balancing decision depends on those variations.
For this inspection cell, each stop needs to be traced to its cause. Reflective parts, uncertain classifications and mechanical faults require different changes. Combining them into one failure rate makes it harder to decide whether to improve perception, tooling or recovery instructions.
- Acceptable units completed, with the part mix and test duration.
- Rejected units and the reason for rejection.
- Operator interventions, time to recover and specialist assistance.
- Changeover time between supported tasks.
Recovery is part of the installation
A missing part, failed grasp or stopped upstream machine should have a defined response. The operator needs to identify the fault, restore a known condition and resume work. Recovery that depends on a developer interpreting logs can become a continuing service obligation.
Assess safety across the whole application: robot, tooling, workpiece, neighbouring machinery and human activity. Component features do not establish the suitability of the assembled cell. Changes to speed, tooling or recovery behaviour may require reassessment against the standards and rules relevant to the installation.
NIST's grasping and manipulation research develops performance metrics and test methods. It is not an approval of a particular application. Keep performance measurement, safety assessment and customer acceptance as distinct responsibilities, and establish who owns each before commissioning.
The next cell tests the business
A second installation exposes the work concealed by the first. Count the engineering hours spent on fixtures, layout, controls, calibration and training. Distinguish changes made through configuration from new mechanical design or software. Bespoke work can support a viable business, provided pricing and delivery capacity reflect it.
Support responsibilities need to be clear at handover. The supplier, integrator and customer should agree ownership of production interfaces, updates, spares and remote support. An unresolved responsibility can keep an otherwise capable cell out of service.
Customer economics include integration, maintenance, consumables and operating support. Time released by automation does not automatically become a cash saving; employees may move to other necessary work, or staffing may remain unchanged. The business case should explain how the observed production result changes capacity, quality or cost under the customer's actual staffing and demand assumptions.
Investment implications
- Count acceptable output over the complete operating period, including interventions and rejected parts.
- Track the work required to install and support the next cell as carefully as the performance of the first.
Sources
Published
