Predicting the Remaining Useful Life (RUL) of turbofan engines is a vital component in the field of prognostics and health management (PHM). Accurate estimation of RUL enables timely maintenance and prevents sudden failures, which is essential for ensuring safety and reducing operational costs. This study introduces a novel hybrid deep learning model referred to as RCBLA (Residual Convolutional Bidirectional Long Short-Term Memory with Attention), designed specifically to handle the temporal and sequential characteristics of engine sensor data. The model architecture combines residual convolutional layers to enhance feature extraction, Temporal Convolutional Networks (TCN) for capturing long-range dependencies, Bidirectional LSTM layers for understanding both past and future contexts, and an attention mechanism to focus on the most critical time steps. Layer normalization is applied to improve convergence and training stability. The proposed model is evaluated using the FD001 and FD003 subsets of the C-MAPSS dataset. A piecewise linear degradation model is adopted, and only sensor measurements are used as input features. A custom learning rate scheduler is applied to optimize model convergence, and a dropout rate of 0.7 and 0.5 is used to prevent overfitting. The results demonstrate the model’s strong predictive capabilities, achieving RMSE values of 14.389 and 12.46 for FD001 and FD003, respectively. The scoring function yields 279.23 and 282.56 for FD001 and FD003, respectively. These outcomes highlight the model’s robustness and suitability for real-world RUL prediction, offering a reliable framework for improving PHM systems in aerospace applications.
The Stochastic Canadian Traveller Problem (SCTP) models adaptive routing on a known graph in which the availability of an edge is revealed only after the traveller reaches an incident vertex, while edge-specific blocking probabilities are known in advance. This paper investigates simulation-based strategies for minimizing the expected travel cost. After summarizing the Optimistic Policy, Hindsight Optimization, Optimistic Rollout, and established UCT-based approaches, the paper formalizes and evaluates two original heuristic variants of Optimistic UCT, UCTO2 and UCTP, which had previously been proposed and implemented in a master's thesis. UCTO2 replaces complete depth-oriented rollout trajectories with progressive tree expansion and batch rollout evaluation of selected nodes. UCTP augments UCTO2 with a pruning mechanism that retains the more promising of alternative tree nodes associated with the same graph vertex. The methods are evaluated on two families of randomly generated graphs with 50 vertices, using ten graphs per family and 500 stochastic realizations per graph. In the reported experiments, UCTP achieved lower mean travel costs than the Optimistic Policy and UCTO while requiring substantially less computation time than the other evaluated UCT variants. HOP and ORO nevertheless attained the lowest overall mean costs. The evidence therefore supports the complete UCTP configuration as a favourable finite-budget alternative within the UCT family, not as a generally superior SCTP policy.
This study analyses and simulates a snake robot's locomotion in complex environments, exploring the impact of serpentine gait parameters like magnitude, shape, and curvature on its motion in linear and circular gaits. The research reveals how small errors in the approximation process can accumulate, significantly affecting the robot's path. Ninety-six scenarios are tested to validate the model and demonstrate that CoppeliaSim's simulations closely match Maple's numerical simulations, especially for longer snakes. The study highlights the minimal role of wheel positions in altering the trajectory and underscores the significance of design parameters and friction in the robot's motion. This comprehensive analysis enhances our understanding of snake robot locomotion and informs their development for diverse applications.
Forecasting intermittent demands is challenging due to their irregular and unpredictable demand pattern. This makes the businesses unprepared for upcoming demands, where the conventional methods often fail to predict the demand occurrence pattern sufficiently. In this paper, we proposed a two-step approach, "UR2CUTE," (Using Repetitively 2 CNN for Unsteady Timeseries Estimation), employing Convolutional Neural Networks (CNNs) specifically designed to handle the unique challenges of intermittent time series. CNNs, known for their effectiveness in capturing spatial and temporal patterns in data, offer a promising area to improve forecast accuracy in predicting time series demand patterns. Our approach presents a combined process for intermittent demand forecasting. A CNN model is initially designed as a binary classifier to determine demand occurrence. Afterward, a distinct CNN model is employed to estimate the magnitude of the demand. This dual-phase approach improves forecasting accuracy in intermittent demands, specifically in predicting the non-demand (Zero-Demand). The suggested approach notably surpasses traditional forecasting techniques, including Croston's method, which is tailored for intermittent demand forecasting. It also outperforms other methods like XGboost, Random Forest, ETR, Prophet, and AutoArima, especially in predicting the lead time demand distribution for sporadic demands. The deployment of dual CNN models facilitates a deeper understanding of intermittent demand dynamics. This, in turn, enhances supply chain management effectiveness and efficiency, offering a robust solution to the complex challenges of intermittent demand forecasting.
Since their introduction, function calls have become a widely used feature within the OpenAI API ecosystem. They reliably connect GPT’s capabilities with external tools and APIs, and they quickly found their way into the other LLMs. The challenge one can encounter is the number of tokens consumed per request since each definition of each function is always sent as an input. We propose a simple solution to effectively decrease the number of tokens by sending only the function corresponding to the user question. The solution is based on saving the functions to the vector database, where we use the similarity score to pick only the functions that need to be sent. We have benchmarked that our solution can decrease the average prompt token consumption by 210% and the average prompt (input) price by 244% vs the default function call. Our solution is not limited to specific LLMs. It can be integrated with any LLM that supports function calls, making it a versatile tool for reducing token consumption. This means that even cheaper models with a high volume of functions can benefit from our solution.
