Organizations: Department of Computer Science and Engineering, Rajshahi University of Engineering and Technology, Rajshahi, Bangladesh · Department of Aeronautical Engineering, Military Institute of Science and Technology, Dhaka 1216, Bangladesh
Accurate Remaining Useful Life (RUL) prediction is a key requirement for effective Prognostics and Health Management (PHM) in safety-critical systems such as aero-engines. Existing deep learning approaches, particularly LSTM-based models, often struggle to generalize across varying operating conditions and are sensitive to noise in multivariate sensor data. To address these challenges, we propose a novel Bidirectional Residual Corrected LSTM (Bi-cLSTM) model for robust RUL estimation. The proposed architecture combines bidirectional temporal modeling with an adaptive residual correction mechanism to iteratively refine sequence representations. In addition, we introduce a condition-aware preprocessing pipeline incorporating regime-based normalization, feature selection, and exponential smoothing to improve robustness under complex operating environments. Extensive experiments on all four subsets of the NASA C-MAPSS dataset demonstrate that the proposed Bi-cLSTM consistently outperforms LSTM-based baselines and achieves competitive state-of-the-art performance, particularly in challenging multi-condition scenarios. These results highlight the effectiveness of combining bidirectional temporal learning with residual correction for reliable RUL prediction.
Figures & tables
Dataset
FD001
FD002
FD003
FD004
Engines in training set
100
260
100
249
Engines in testing set
100
259
100
248
Training trajectories
17,731
48,558
21,120
56,815
Testing trajectories
100
259
100
248
Max/min cycle for train
362/128
378/128
525/145
543/128
Max/min cycle for test
303/31
367/21
475/38
486/19
TABLE I : Each Sub-Unit of the C-MAPSS Dataset
Sensor Parameter
Description with Units
T2
Total Temperature in fan inlet (°R)
T24
Total Temperature at LPC outlet (°R)
T30
Total Temperature at HPC outlet (°R)
T50
Total Temperature at LPT outlet (°R)
P2
Pressure at fan inlet (psia)
P15
Total pressure in bypass-duct (psia)
TABLE II : Output Parameters of the C-MAPSS Turbofan Engine Datasets
Fig. 1 : Proposed Bi-cLSTM framework with a residual Corrector module. Input features are projected, processed by bidirectional LSTMs, refined through adaptive correction, and mapped to Remaining Useful Life (RUL) predictions.
Architecture
RMSE
MAE
R2
LSTM
27.5375
0.1607
0.5276
cLSTM
18.9375
0.1089
0.7766
Bi-LSTM
20.2125
0.1194
0.7454
Bi-cLSTM
17.5125
0.1066
0.8089
TABLE III : RUL prediction performance (RMSE, MAE in units; R2 dimensionless) on FD001 Sub-Unit
Dataset
2 Blocks
4 Blocks
6 Blocks
8 Blocks
10 Blocks
FD001
19.59
17.51
18.11
21.81
21.91
FD002
14.73
13.96
14.10
14.12
14.61
FD003
24.61
21.85
22.98
22.98
21.38
FD004
14.74
14.25
15.44
14.63
15.74
TABLE IV : Bi_cLSTM ablation across number of blocks using RMSE (lower is better).
Method
Year
Pre-processing steps
FD001
FD002
FD003
FD004
CNN [ 17 ]
2016
Data normalization; RUL target
18.44
30.29
19.81
29.15
LSTM [ 18 ]
2017
Data normalization; RUL target
16.14
24.49
16.18
28.17
BiLSTM [ 19 ]
2018
Feature sel.; norm; RUL target
13.65
23.18
13.74
24.86
CNN+LSTM [ 20 ]
2019
Var. threshold; norm; HI
16.16
20.44
17.12
23.25
Multi-head CNN+LSTM [ 21 ]
2020
Feature sel.; RUL target
12.19
19.93
12.85
22.89
CNN+LSTM+BiLSTM [ 5 ]
2020
Corr. analysis; min–max; RUL target
10.41
–
–
–
TABLE V : Chronological RMSE comparison on C - MAPSS (FD001–FD004). RMSE in cycles (lower is better).