Prediction of Performance Degradation in Safety-Critical Flight Software Using NASA SimuPy-Flight and Random Forest

This paper describes a simulation-based model to identify execution level performance degradations in safety critical flight software using synthetic telemetry that was generated with NASA’s SimuPy-Flight simulation environment. Anomalies such as CPU stalls, sensor-pipeline lag, actuator-response delay, command-queue congestion, scheduling jitter, and telemetry dropouts are also fed in to simulated sequences of attitude-propagation to provide realistic timing disturbances. Telemetry is divided into fixed time windows and converted into statistical and time related descriptors. These window characteristics are used to classify a Random Forest classifier, which is used to differentiate between nominal and degraded execution segments. The model has an accuracy of 0.8686, precision of 0.9052, recall of 0.8235, and an F1-score of 0.8624 in held-out test data on using 100 simulation runs. The analysis of feature-importance shows the presence of body-rate dynamics and quaternion derivatives, which are the most significant factors that indicate timing degradation. The framework offers a low-weight, interpretable profiling feature to characterize flight- software behavior under timing stress to allow early detection of execution-level degradations before hardware-in-the-loop or system-integration testing. The methodology, as it is fully based on simulation, allows repeatable experiments and eliminates reliance on proprietary mission telemetry and introduces a viable route to the implementation of ML-based performance diagnostics into aerospace software-development processes.