Based on a typical industrial dual-sensor redundancy comparison scenario (taking injection molding hot runner pressure detection as an example; other scenarios can be scaled up proportionally), here's a set of practically validated experimental parameter settings you can directly apply:
I. Basic Experimental General Parameters
|
Parameter Item |
Suggested Setting Value |
Explanation |
|
Test Cycle |
≥14 days per algorithm (covering ≥5000 production cycles) |
Ensures sufficient sample size and excludes accidental fluctuations |
|
Long-Term Stability Test Duration |
≥90 days of continuous operation |
Verifies the adaptability of the adaptive mechanism to aging |
|
Sampling Frequency |
10Hz (samples every 100ms) |
Meets the detection needs of dynamic industrial processes, and has appropriate computational power requirements |
|
Filter Window Size |
Median filtering: 3 points; Moving average filtering: 10 points |
Balances noise reduction and response speed; large windows are not recommended |
|
Alarm Delay Threshold |
10 seconds |
Filters most instantaneous interference without delaying real fault alarms |
|
EWMA Smoothing Coefficient α |
0.01~0.05 (0.02 recommended) |
Slowly follows aging, preventing baseline disruption from single anomalies. |
Adaptive Update Cycle: Weekly (executed during equipment maintenance windows) Balances computing power usage and update timeliness.
II. Deviation Threshold Basic Parameters (Taking a 0~10MPa injection molding pressure sensor as an example)
|
Parameter Item |
Recommended Setting Value |
Description |
|
Initial Basic Threshold |
0.05%FS (corresponding to 0.005MPa) |
A reasonable initial value for the standard accuracy level of industrial pressure sensors. |
|
Threshold Coefficients for Each Stage |
- Injection Stage: 2.5 times - Mold Opening and Closing Stage: 1.5 times - Holding Pressure Stage: 1.0 times - Cooling Stage: 0.8 times |
Matches fluctuations in operating conditions at different stages, dynamically balancing sensitivity and false alarm rate. |
|
Minimum Threshold Limit |
Initial Basic Threshold × 50% |
Prevents the false alarm rate from rising due to excessively low thresholds after adaptive updates. |
III. Fault Simulation Experiment Parameters (Same as the above 0~10MPa sensor scenario)
|
Simulated Fault Type |
Parameter Setting Recommendation |
|
Instantaneous Interference Drift Single sensor offset accumulation |
0.1%FS (0.01MPa), duration: recovers after 1 second. Simulates injection impact and electrical interference scenarios. |
|
Continuous slow aging: Single sensor daily offset accumulation |
0.01%FS (0.001MPa), stops when it reaches 0.1%FS (0.01MPa). Simulates natural aging drift of the sensor. |
|
Step-like sudden drift: Single sensor continuous offset accumulation |
0.2%FS (0.02MPa). Simulates sudden failure drift of the sensor. |
IV. Statistical significance test parameters: The significance level is set to p<0.05: As long as the chi-square test result satisfies p<0.05, the algorithm difference can be determined to be real and effective, and the engineering scenario is rigorous enough.
If the false alarm rate decreases by ≥60%, a chi-square test is not required; the superior performance can be directly determined.
Parameter Adjustment Tips for Different Scenarios:
For temperature detection scenarios: the initial base threshold can be relaxed to 0.1%FS, as temperature fluctuations are greater than pressure fluctuations;
For high-frequency dynamic processes: the alarm delay can be shortened to 5 seconds to ensure response speed;
For slow-variable industrial processes: the moving average window can be enlarged to 20 points for better filtering.
These parameters have been validated in industrial batches and can be directly applied to obtain reliable experimental results without the need for repeated trial and error.
