Optimizing the performance of the dual-sensor redundancy comparison algorithm hinges on upgrading from "static comparison" to "dynamic intelligent fusion," focusing on resolving false alarms, missed alarms, and accuracy fluctuations. This is implemented in four steps:
1. Introduce multi-stage dynamic thresholds to adapt to the injection molding process
Injection molding involves drastic pressure changes, and fixed thresholds are prone to false alarms. Automatic switching based on process stages is necessary:
Injection/Holding Pressure Stage (drastic pressure fluctuations): Relax the threshold to 0.2%-0.3%FS and enable a 5-10 second delay for judgment to filter instantaneous impacts;
Cooling/Mold Opening Stage (pressure stabilizes at 0): Tighten the threshold to 0.05%FS to accurately capture zero-point drift;
The system reads the injection molding machine's process signals in real time and automatically matches the corresponding threshold, reducing false alarms by more than 80%.
2. Upgrade the filtering algorithm to eliminate noise interference.
Avoid direct comparison with raw data; preprocess first:
Use a combination of moving average filtering and median filtering: first remove sudden spikes in noise, then smooth random fluctuations to extract the true trend;
It is recommended to set the sampling window to 20-30 points to ensure response speed while effectively suppressing high-frequency noise, making the comparison benchmark more stable.
3. Establish a weighted fusion model to improve output accuracy.
When both sensors are functioning normally, simple averaging is not the optimal solution:
Assign weights based on historical stability: give higher weights to sensors with smaller long-term drift and lower noise (e.g., 0.6:0.4);
Dynamically update weights: re-evaluate the stability of both sensors every 1000 simulations, automatically adjusting the weight ratio to ensure the final output always closely matches the higher-precision side, improving overall accuracy by 10%-15%.
4. Enhance trend prediction logic for early warning.
In addition to monitoring current deviations, also consider the rate of change:
Monitor the first derivative of the deviation (rate of change): If the deviation continues to increase unidirectionally within 24 hours, even if it hasn't exceeded a threshold, mark it as "potentially aging" and remind users to perform planned maintenance.
This transforms passive alarms into proactive prevention, avoiding production interruptions caused by sudden failures.
This optimization scheme allows the algorithm to maintain high reliability under complex operating conditions while maximizing the accuracy advantages of dual sensors.
