Table 7 Comprehensive comparison with advanced fault detection and optimization techniques.
Technique | Fault detection accuracy (%) | Computational complexity | Real-time performance | Advantages | Limitations |
|---|---|---|---|---|---|
Traditional Methods (PCA, SVM) | 70–85% | Low | Low | Easy to implement, less data needed | Struggles with high-dimensional or nonlinear data |
CNN | 85–95% | High | Moderate (Batch processing) | Powerful for spatial data, good at anomaly detection | Needs large datasets, computationally expensive |
LSTM | 85–95% | High | Moderate to High (Sequential) | Excellent for sequential and temporal data, robust | High computational cost, large datasets needed |
GRU | 85–95% | Moderate | Moderate (Sequential) | More efficient than LSTM, comparable performance | Not suitable for very complex temporal patterns |
SVM | 75–85% | Low to Moderate | Low | Effective for smaller datasets, easy to implement | Struggles with large or high-dimensional datasets |
Autoencoders (AE) | 75–90% | Moderate to High | Low to Moderate (Batch) | Good for detecting anomalies, reduces dimensionality | Sensitive to noise, requires large datasets |
DT + Deep Learning | 90–98% | Very High | High (Real-time) | Real-time monitoring and fault detection, highly adaptable | High computational cost, needs real-time data and sensors |
Genetic Algorithms (GA) | 75–90% | High | Moderate | Versatile, good for complex, non-linear systems | Computationally expensive, sensitive to parameter settings |
Particle Swarm Optimization (PSO) | 70–85% | Moderate to High | Moderate to Low | Suitable for multi-dimensional optimization problems | Can get trapped in local minima, requires tuning |
