Could you provide specific suggestions for experimental parameter settings?

Here is a set of practically validated experimental parameter settings that you can immediately implement, based on a typical industrial dual-sensor redundancy comparison scenario (using injection molding hot runner pressure detection as an example; alternative scenarios can be scaled up proportionally):

 

I. Fundamental General Experimental Parameters

The parameter item

Recommended Setting Value

Justification

Test Period

Each algorithm takes at least 14 days, spanning at least 5000 production cycles.

guaranties an adequate sample size and eliminates unintentional variations

Test Duration for Long-Term Stability

≥90 days of nonstop operation

confirms the adaptive mechanism's capacity to adjust to age

Frequency of Sampling

10Hz (every 100 milliseconds)

satisfies the detection requirements of dynamic industrial processes and requires the right amount of processing power.

Size of the Filter Window

Three points for median filtering and ten points for moving average filtering

balances response time with noise reduction; big windows are not advised.

Threshold for Alarm Delay

Ten seconds

eliminates the majority of instantaneous interference without postponing actual problem alerts

Coefficient of EWMA Smoothing α

0.01–0.05 (ideally 0.02)

prevents single abnormalities from upsetting the baseline by gradually following aging.

Weekly Adaptive Update Cycle (performed during equipment maintenance times) strikes a balance between update timeliness and processing power consumption.

 

II. Deviation Threshold Basic Parameters (using an injection molding pressure sensor operating at 0–10 MPa as an example)

The parameter item

Suggested Setting Value

An explanation

First Basic Threshold

0.05%FS, or 0.005MPa

a fair starting point for industrial pressure sensors' typical accuracy level.

Each Stage's Threshold Coefficients
2.5 times for injection; 1.5 times for mold opening and closing; 1.0 times for holding pressure; and 0.8 times for cooling

dynamically balances sensitivity and false alarm rate by matching changes in operating conditions at various stages.

Minimum Limit on Threshold

First Basic Threshold x 50%

stops the rate of false alarms from increasing as a result of extremely low thresholds following adaptive updates.

 

III. Fault Simulation Experiment Parameters (Identical to the 0~10MPa sensor case mentioned above)

Type of Simulated Fault

Recommendation for Parameter Setting

Accumulation of Instantaneous Interference Drift Single Sensor Offset

0.1%FS (0.01MPa), duration: recovers in one second. Electrical interference and injection impact scenarios are simulated.

Slow aging over time: daily offset accumulation from a single sensor

When it hits 0.1%FS (0.01MPa), it stops at 0.01%FS (0.001MPa). mimics the sensor's normal aging drift.

Step-like abrupt drift: Continuous offset accumulation from a single sensor

0.02MPa, or 0.2%FS. simulates the sensor's abrupt failure drift.

 

IV. Statistical significance test parameters: The chi-square test result must satisfy p<0.05 in order for the algorithm difference to be deemed actual and effective, and the engineering scenario must be sufficiently stringent. The significance threshold is set at p<0.05.

A chi-square test is not necessary if the false alarm rate drops by ≥60%; the enhanced performance can be ascertained directly.

Advice on Changing Parameters in Various Situations:

Since temperature fluctuations are larger than pressure changes, the initial base threshold for temperature detection scenarios can be lowered to 0.1%FS;

To guaranty response time for high-frequency dynamic processes, the alarm delay can be reduced to five seconds;

The moving average window can be expanded to 20 points for improved filtering in slow-variable industrial processes.

These parameters can be used directly to get trustworthy experimental findings without the need for repeated trial and error because they have been verified in industrial quantities.