Common dual-sensor redundancy comparison algorithms can be divided into three categories: basic introductory type, optimized advanced type, and intelligent adaptive type, adapting to scenarios of varying complexity. Their specific characteristics are as follows:
1. Basic Introductory Type: Fixed Threshold Direct Comparison Method
The most classic introductory algorithm, the simplest to implement, suitable for simple retrofits in small and medium-sized factories:
Core Logic: Set a fixed deviation threshold; if the difference between the outputs of the two sensors exceeds the threshold, an alarm is triggered directly. No other logic is required.
Advantages: Minimal development effort; can be directly written to existing PLCs without complex configuration.
Disadvantages: Extremely high false alarm rate; cannot adapt to process fluctuations and sensor aging; only suitable for scenarios with low reliability requirements; currently being phased out.
2. Optimized Advanced Type: Staged Dynamic Threshold Comparison Method
The current mainstream algorithm in industrial settings, suitable for periodic production scenarios such as injection molding:
Core Logic: Divide different thresholds according to process stages; relax the threshold during dynamic stages such as injection/start/stop, tighten the threshold during steady-state production stages, and add delayed confirmation and dual filtering to filter interference.
Advantages: False alarm rate reduced by over 70% compared to a fixed threshold; simple to implement, requires no complex computing power, and can be configured by ordinary engineers.
Applicable Scenarios: Most redundancy detection scenarios in injection molding and process industries; currently the most widely used mainstream solution.
3. Advanced Optimization: 3σ Adaptive Threshold Update Method
Adds automatic adaptation capability to the dynamic threshold, further reducing long-term false alarms:
Core Logic: Historical deviation data is collected weekly, and the normal fluctuation range of the deviation is calculated using the 3σ principle. The current threshold is automatically updated, slowly adapting to drift caused by sensor aging.
Advantages: No need for manual periodic threshold adjustment; good long-term operational stability; false alarm rate reduced by another 30% compared to a fixed dynamic threshold.
Applicable Scenarios: Continuously operating mass production lines where reduced manual maintenance is required.
4. Intelligent Adaptive Type: EWMA Exponentially Weighted Baseline Drift Compensation Method
An intelligent optimization algorithm for high-end scenarios, highly effective in handling slow sensor aging:
Core Logic: Uses Exponentially Weighted Moving Average (EWMA) to update the baseline deviation between two sensors in real time, automatically compensating for the aging sensor's output. An alarm is only triggered when the deviation exceeds the compensation range.
Advantages: Automatically follows slow aging, extremely low long-term false alarm rate, extends sensor replacement cycle, and reduces maintenance costs;
Disadvantages: Requires some computation, but existing PLCs can fully support it, and implementation is not difficult.
5. High-End Intelligent Type: Machine Learning Anomaly Detection Method
A solution for fully automated high-end production lines:
Core Logic: Trains models such as Isolation Forest and LSTM using historical normal data, automatically learning the normal deviation range under different processes, and identifying abnormal deviations in real time.
Advantages: Highly adaptable, lowest false alarm rate in complex and variable processes;
Disadvantages: Requires a large amount of historical data for training, slightly higher computing power requirements, low cost-effectiveness for small and medium-sized factories, currently only used in high-end medical and aerospace scenarios.
