Computation Time Research Articles

Research on path smoothing optimisation based on improved RRT-Connect algorithm and third-order Bezier curve

Aiming at the deficiencies of the original RRT-Connect path planning algorithm in dealing with obstacle avoidance, planning efficiency and path smoothing in static environments, an improved path optimisation method is proposed by fusing the RRT-Connect path planning algorithm with the greedy search strategy and using a third-order Bezier curve for path smoothing. Firstly, a greedy algorithm is added to the path planning process of the original RRT-Connect algorithm to guide the search direction, so as to make the path search more goal oriented, reduce the time of path planning and improve the efficiency. Secondly, the generated paths are smoothed with third-order Bezier curves to ensure the generation of smooth paths with continuous curvature and improve the quality of the planned paths. Finally, the improved RRT-Connect smoothing optimisation algorithm is simulated with the original RRT-Connect, A* and Dijkstra algorithms in different simulation environments to check the performance of the optimised algorithm. The results show that the success rate of obstacle avoidance in planning paths of the optimised algorithms are all 100%, and compared with the original RRT-Connect algorithm, the improved RRT-Connect smoothing optimisation algorithm has an overall reduction of 12.02% in planning path length, an overall reduction of 40.22% in computation time and an overall improvement of 69.15% in path smoothness. Compared to the A* algorithm, which is based on the graph search shortest path planning algorithm and takes the shortest time, there is an overall reduction of 62.46% in computation time. The improved RRT-Connect smoothing optimisation algorithm has shorter path planning lengths, smoother paths and shortest computation time in different simulation environments. Meanwhile, the stability of the optimisation algorithm is effectively improved in different environments and different running times.

Land layout optimisation for virtual land sales in the metaverse: two-dimensional assortment problem

With the growing popularity of the metaverse, the virtual real estate in the metaverse has generated immense investment enthusiasm. Considering the virtual land sale mode allowing the potential buyers to request the purchase of rectangular land parcels of a specified size, this study investigates how to lay out the requested land parcels for minimising the size of the required rectangular open land, which can be formulated completely as the two-dimensional assortment problem (2DAP). Due to the strong NP-hardness of the 2DAP, an effective and efficient heuristic solution approach named binary adjoining algorithm (BAA) is presented for tackling the 2DAPs in large scales. The conducted computational experiments show that the BAA can outperform the state-of-the-art piecewise-linearisation mixed integer linear programming model as well as four existing advanced metaheuristic techniques designed for the 2DAP, in both solution quality and computational time, on the small-size instances. The superiority of the BAA over either foregoing reference method on the large-size instances with up to 60 requested land parcels is also demonstrated.

A Highly Hardware Efficient ML-KEM Accelerator with Optimised Architectural Layers

The Module-Lattice-Based Key encapsulation Mechanism (ML-KEM) scheme, which is currently being standardized, is a quantum attack resistant KEM that is based on CRYSTALS-Kyber. CRYSTALS-Kyber is the only Public-key Encryption (PKE)/ KEM scheme selected in the first set of successful candidates as part of the NIST initiated Post-Quantum Cryptography (PQC) process. ML-KEM scheme includes three different security levels, namely security level 1, 3, and 5. In this research, we propose a highly area-time efficient hardware ML-KEM architecture. The architecture comprises three computational layers. The first layer comprises a hash and sampling module; the second layer includes a number theoretic transform (NTT), its inverse (INTT) and a point-wise multiplication (PWM) module; and the third layer comprises addition, compressing and encoding. Intra-layer pipelining and out-of-layer scheduling ensures that either layer 1 or layer 2 operate in the shortest time. In the reduction module, we propose a novel hybrid architecture to obtain the final result within 2 cycles with low area consumption. In the NTT module, the PWM pipelining method is modified and an optimised iterative FIFO access method is adopted to reduce the size of FIFO units by 55% over previous research. Look-up tables are also used to replace the first-stage of the NTT to reduce 8 cycles. Furthermore, the memory unit uses only FIFOs, the size are optimised based on the requirements of the most resource-intensive function in ML-KEM (ML-KEM.CPA.Dec). The results show that the proposed architecture 48.2%, 41.2%, and 78.1% reduction in computational time in comparison to previous work for security level 1, 3 and 5, respectively. In addition, the area of proposed optimised ML-KEM designs is reduced by 73%, 70%, 76% and resulting in an improved area-time (AT) product of 15.8%, 10.7%, and 11.3%, for the Level 1, 3 and 5 security levels respectively, compared with state-of-the-art designs.

Rapid Analysis of DEER Signals Including Short Distances.

Double electron electron resonance (DEER) spectroscopy is an important technique to measure distance distributions P(r) for studying protein structures and protein-protein interactions. DEER data analysis can at times become challenging due to the lack of a detailed analytical signal expression or numerical methods with rapid computation time. We have derived an analytical expression κFULL, which includes both the pseudo-secular dipolar coupling (PSDC) and the finite pulse effects, especially important for shorter distances. Analyses of experiments by κFULL yield accurate and consistent P(r) values for three DEER nitroxide-rulers with distances (rAVG) in the range of 15 to 32 Å, while the current standard analysis produces erroneous results for rAVG < 20 Å. Computation times for deriving P(r) vary between 1 min and 4 min, which is usually much shorter than previous methods that include pseudo-secular and other effects. The expression can be applied to all types of DEER spin probes with little or no modifications.

Breast cancer image classification by using HCNN and LeNet5

Medical data from many sectors has greatly increased during the last 10 years. One of the main causes of death and illness among women worldwide is breast cancer. Breast cancer (BC) is one of the most common cancers in women; the death rate is high and holds the second position next to lung cancer. Breast cancer develops when cells in the mammary glands and the ducts that transfer milk to the nipple grow out of control. This is the first step in the progression of the disease. However, the time complexity of the available techniques is immense due to the many processes. Additionally, the current research attempts to improve the computation time involved in the detection process. Therefore, an effective hybrid deep learning model is introduced to improve the prediction performance and reduce the time consumption compared to the machine learning model. The breast cancer dataset, obtained from Kaggle, is used as the input data. A Wiener filter preprocessing technique is applied to enhance the image quality, with an active Wiener filter employed for this purpose. The segmentation step is achieved using a Modified Watershed Algorithm, which isolates the region of interest within the images. Finally, classification is performed using a hybrid deep learning model. This model combines a Convolutional Neural Network (CNN) with an Enhanced Recurrent Neural Network (ERNN), leveraging the strengths of both architectures. According to experimental results, the proposed Hybrid Convolutional Neural Network (HCNN) model achieves an accuracy of 96.12%, a precision of 96.99%, a recall of 97.52%, and an F-measure of 97.25%, outperforming other existing models.

Computationally efficient collimator-detector response compensation in high energy SPECT using 1D convolutions and rotations.

Modeling of the collimator-detector response (CDR) in SPECT reconstruction enables improved resolution and accuracy, and is thus important for quantitative imaging applications such as dosimetry. The implementation of CDR modeling, however, can become a computational bottleneck when there are substantial components of septal penetration and scatter in the acquired data, since a direct convolution-based approach requires large 2D kernels. This work proposes a 1D convolution and rotation-based CDR model that reduces reconstruction times but maintains consistency with models that employ 2D convolutions. To enable open-source development and use of these models in image reconstruction, we release a SPECTPSFToolbox repository for the PyTomography project on GitHub. A 1D/rotation-based CDR model was formulated and subsequently fit to Monte Carlo point source data representative of Lu-177, I-131, and Ac-225 imaging. Computation times of (i) the proposed 1D/rotation-based model and (ii) a traditional model that uses 2D convolutions were compared for typical SPECT matrix sizes. Both CDR models were then used in the reconstruction of Monte Carlo, physical phantom, and patient data; the models were compared by quantifying total counts in hot regions of interest (ROIs) and activity contrast between hot ROIs and background regions. For typical matrix sizes in SPECT reconstruction, application of the 1D/rotation-based model provides a two-fold computational speed-up over the 2D model when running on GPU. Only small differences between the 1D/rotation-based and 2D models (order of 1%) were obtained for count and contrast quantification in select ROIs. A technique for CDR modeling in SPECT was proposed that (i) significantly speeds up reconstruction times, and (ii) yields nearly identical reconstructions to traditional 2D convolution based CDR techniques. The released toolbox will permit open-source development of similar models for different isotopes and collimators.

Enhancing generalizability of data-driven urban flood models by incorporating contextual information

Abstract. Fast urban pluvial flood models are necessary for a range of applications, such as near real-time flood nowcasting or processing large rainfall ensembles for uncertainty analysis. Data-driven models can help overcome the long computational time of traditional flood simulation models, and the state-of-the-art models have shown promising accuracy. Yet the lack of generalizability of data-driven urban pluvial flood models to both unseen rainfall and distinctively different terrain, at the fine resolution required for urban flood mapping, still limits their application. These models usually adopt a patch-based framework to overcome multiple bottlenecks, such as data availability and computational and memory constraints. However, this approach does not incorporate contextual information of the terrain surrounding the small image patch (typically 256 m×256 m). We propose a new deep-learning model that maintains the high-resolution information of the local patch and incorporates a larger surrounding area to increase the visual field of the model with the aim of enhancing the generalizability of data-driven urban pluvial flood models. We trained and tested the model in the city of Zurich (Switzerland), at a spatial resolution of 1 m, for 1 h rainfall events at 5 min temporal resolution. We demonstrate that our model can faithfully represent flood depths for a wide range of rainfall events, with peak rainfall intensities ranging from 42.5 to 161.4 mm h−1. Then, we assessed the model's terrain generalizability in distinct urban settings, namely, Lucerne (Switzerland) and Singapore. The model accurately identifies locations of water accumulation, which constitutes an improvement compared to other deep-learning models. Using transfer learning, the model was successfully retrained in the new cities, requiring only a single rainfall event to adapt the model to new terrains while preserving adaptability across diverse rainfall conditions. Our results indicate that by incorporating contextual terrain information into the local patches, our proposed model effectively simulates high-resolution urban pluvial flood maps, demonstrating applicability across varied terrains and rainfall events.

Assessment of evolutionary multi-objective optimisation algorithms for thermoplastic injection moulding in aluminium moulds

In order for a plastic part manufactured by the injection moulding process to be of the desired quality, it is necessary to maintain control of the process parameters. To increase productivity and reduce defects in parts, various studies have been carried out to propose and evaluate methods for determining injection parameters, which are based on the use of experimental planning techniques as well as the application of optimisation algorithms. This work proposes an experimental design with 6 injection parameters (mould temperature, melt temperature, injection time, percentage of volume filled, packing time and packing pressure) to determine the initial search space for the NSGA-II and MOEA/D optimisation algorithms in relation to the use of aluminium moulds. This procedure was used via a computer code described in the PYTHON programming language to determine the appropriate injection parameters for minimising part warpage and injection pressure. The results showed that an increase in compaction pressure is associated with a reduction in warpage and vice versa. It was observed that the NSGA-II algorithm showed a greater diversity of points on the approximate Pareto frontier compared to the MOEA/D algorithm. In the case of the proposed study, MOEA/D presented a more viable solution, determining a lower packing pressure, less warping of the part and a shorter cycle time. The NSGA-II algorithm required around 4 hours more computational time than MOEA/D.

Improved Active Vector Optimization Strategy‐based Model Predictive Torque Control for Matrix Converter‐Fed PMSM With Reduced Torque and Flux Ripples

ABSTRACTThe Matrix Converter‐fed Finite Control Set‐Model Predictive Torque Control (FCS‐MPTC) technique has superior potentiality in the control of electrical drives due to its availability of increased number of voltage vectors with compact topology. On the other side, the FCS‐MPTC technique suffers a major drawback of high computational complexity and processing time. This major drawback has degraded the attention of the FCS‐MPTC technique in high‐power industrial applications because it needs expensive and high‐rating processors for the functioning of the control algorithm. In this regard, this paper proposes an efficient active vector optimization concept by formulating a switching table for the optimal voltage vector selection. This approach reduces the available 18 active vectors to three, thereby decreasing the execution time of the cost function (CF). In addition, to mitigate the high torque and flux ripple problems associated with the existing FCS‐MPTC method, a modified switching table is proposed. The effectiveness of the proposed approach is evaluated by conducting various test loadings using the MATLAB Simulink, and the results are further validated in real time using the OPAL‐RT laboratory setup.

Search anything: segmentation-based similarity search via region prompts

AbstractSearch Anything is presented, a novel approach to perform similarity search in images. In contrast to other approaches to image similarity search, Search Anything enables users to utilize point, box, and text prompts to search for similar regions in a set of images. The region selected by a prompt is automatically segmented, and a binary feature vector is extracted. This feature vector is then used as a query for an image region index, and the images that contain the corresponding regions are returned. Search Anything is trained in a self-supervised manner on mask features extracted by the FastSAM foundation model and semantic features for masked image regions extracted by the CLIP foundation model to learn binary hash code representations for image regions. By coupling these two foundation models, images can be indexed and searched at a more fine-grained level than finding only entire similar images. Experiments on several datasets from different domains in a zero-shot setting demonstrate the benefits of Search Anything as a versatile region-based similarity search approach for images. The efficacy of the approach is further supported by qualitative results. Ablation studies are performed to evaluate how the proposed combination of semantic features and segmentation features together with masking improves the performance of Search Anything over the baseline using CLIP features alone. For large regions, relative improvements of up to 9.87% in mean average precision are achieved. Furthermore, considering context is beneficial for searching small image regions; a context of 3 times an object’s bounding box gives the best results. Finally, we measure computation time and determine storage requirements.

Study on the Uncertainty of Input Variables in Seismic Fragility Curves Based on the Number of Ground Motions

Seismic fragility curves, derived from ground motion data, are essential tools for predicting and assessing potential damage to structures during earthquakes. Seismic fragility curves are vital for assessing the structural behavior of buildings and establishing disaster response criteria when an earthquake occurs. We performed an incremental dynamic analysis based on 400 ground motion data. We sampled various sets of ground motions (10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 150, 200, 250, 300, and 350) and derived seismic fragility curves for three performance criteria, based on inter-story drift, by conducting 100,000 simulations for two steel frame structures each (6-story and 13-story). Fewer ground motions increase the uncertainty of the seismic fragility curve, distorting the results. Conversely, increasing the number of ground motions improves the reliability of the input variables and enhances the consistency of the results. The median and the logarithmic standard deviation for both structures converged toward the reference values when 30 or more ground motions were used. Similar results were observed when ≥50 ground motions were used. Specifically, more ground motions corresponded with a lower uncertainty in deriving the input variables for the seismic fragility curve, improving the reliability of the results. In conclusion, the number of ground motions used is directly related to the computational time for numerical analysis when deriving seismic fragility curves. Therefore, considering an appropriate number of ground motions is crucial to enhancing the reliability of the input variables used in evaluating the structural performance.

Limitations of calculating theoretical solutions for closed BCMP queueing networks and verification of alternative theoretical values by parallel simulation.

This study applied a closed BCMP queueing network to a real-world model, examining the limitations of the theoretical solution and the possibility of replacing theoretical values with those from parallel simulation. Parallel computing was applied to mean value analysis (MVA). We first obtained computational and theoretical values by varying the number of nodes from 33 to 300 and customers from 250 to 1500 in a system with three customer classes. The computation time increased proportionally with the number of nodes but exponentially with the number of customers, reaching 146,798.86 seconds for 33 nodes, 3 customer classes, and 1500 customers. We then considered a system with more customer classes; due to the greater computational burden, we proposed addressing this problem with simulation. By using a large-scale computing environment (a supercomputer), it was possible to obtain the theoretical solutions for up to three customer classes and verify the simulation accuracy. The parallel simulations' performance-evaluation indices, such as the average number of people in the system, converged to the theoretical values within an acceptable error range after 100,000 simulation hours for systems with four or more customer classes. These results demonstrate that the proposed parallel simulation approach can serve as an accurate and computationally efficient alternative to theoretical solutions for large-scale closed BCMP queueing networks.

Optimisasi Model Backpropagation untuk Meningkatkan Deteksi Kejang Epilepsi pada Sinyal Electroencephalogram

Epilepsy is a chronic neurological disorder characterized by recurrent seizures caused by abnormal electrical activity in the brain. Fast and accurate seizure detection is crucial to support medical intervention and improve patients' quality of life. Currently, Electroencephalogram (EEG) signals are widely used to diagnose epilepsy as they record brain electrical activity in real-time. However, manual analysis of EEG signals requires time and precision, necessitating a more effective automated solution. This study aims to optimize the Backpropagation model for detecting epileptic seizures using EEG data. The research involved collaboration between Telkom University, Sumber Waras Hospital, and the University of Bonn. The EEG data collected was processed through Discrete Cosine Transform (DCT) to extract important features before being used to train the artificial neural network (ANN) model. The model was trained and tested using varying numbers of epochs to measure its accuracy. The results show that the Backpropagation model achieved optimal accuracy of 91.15% at 100 epochs and increased to 93.05% at 200 epochs. Although accuracy improved with more epochs, the longer computational time posed a risk of overfitting. This research demonstrates that the Backpropagation algorithm can be optimized to detect epileptic seizures accurately and efficiently. The implication for Sumber Waras Hospital is that this model can be implemented in EEG monitoring systems to detect seizures in real-time, supporting faster medical intervention and reducing reliance on manual analysis. Thus, this study contributes to providing a more efficient diagnostic solution and enhancing healthcare services for epilepsy patients.

An optimised delay multiply and sum ultrasonic imaging algorithm based on sparse arrays and parallel computing

ABSTRACT The Total Focusing Method (TFM) focuses pixels using the Delay and Sum (DAS) beamforming technique, which relies solely on the temporal information of the full matrix capturing dataset while ignoring its spatial information, and the image resolution and contrast achievable with TFM are limited. In this work, a parallel sparse delay multiply and sum (PSDMAS) focusing imaging algorithm based on sparse arrays and parallel computing is proposed to improve contrast resolution and imaging efficiency. A sparse array optimisation method is applied to reduce the amount of data. A ratio of main-lobe width and side-lobe peak was constructed as the fitness function and a genetic algorithm was used to find the optimal solution for the array arrangement. Delay Multiply and Sum (DMAS) was employed to enhance the spatial coherence and suppress the clutter artefacts. Parallel computing strategies were implemented to improve imaging efficiency. To validate the effectiveness of the algorithm, we processed the full matrix data collected from simulations and experiments using PSDMAS, the imaging results of the PSDMAS provided a considerable improvement in Array Performance Indicator (API), and better lateral spatial resolution was also achieved. The computation time of PSDMAS was reduced by 99.9% compared to conventional DMAS.

ACCELERATED SUBSPACE ITERATION METHOD FOR SIMULATING VARYING EIGENPAIRS OF THIN-WALLED WORKPIECES

This paper presents an innovative accelerated subspace iteration method tailored for simulating varying eigenpairs of thin-walled workpieces. The proposed approach significantly boosts computational efficiency by selecting optimal start iteration vectors for a given cutter location. The optimal start iteration vectors are generated based on the previously calculated eigenvectors from the preceding cutter location. A numerical example illustrates the efficiency enhancement achieved by the proposed method, showcasing a notable reduction in computation time. Furthermore, to evaluate the numerical accuracy of the accelerated subspace iteration method, comprehensive comparisons are conducted with results obtained from Abaqus simulations. The outcomes affirm the numerical accuracy of the proposed approach.

Nonlinear Dynamic Analysis Using Harmonic Balance Method

There is a high demand for nonlinear structural dynamics in implicit Finite Element Analysis (FEA). Although such methods are available, there are several obstacles to use them daily. One is their extreme and unpredictable computation time, which makes it often impossible to get results in time. Another point is the restriction of the methods to the time domain, which is often in contrary to the usual design rules based on frequency domain results. The Harmonic Balance Method (HBM) is a solution for an important sub-class of analysis cases, which resolves the two mentioned obstacles. As a starting point, we define HBM as a frequency response analysis with local nonlinearities. This allows to solve contact problems or mounting problems with nonlinear force-deflection curves. The primary results of HBM are in frequency domain. For all calculated frequencies, a solution in time domain is also available for a periodic response. A simplified radiator is used as industrial example. To prove the validity of HBM, a comparison with a linear frequency response analysis is performed, which shows same results. Then, rubber bushes and contact are added to the model as nonlinearities. Key results of stress and fatigue are presented, and the computation times are analysed to demonstrate the feasibility of the HBM implementation for applications in research and industry. All simulations are performed with the FEA software PERMAS, which contains the HBM among many other analysis methods in structural dynamics.

A fingerprinting structure model for Arabic document plagiarism detection

Plagiarism, which is a significant problem in the academic world worldwide, is particularly challenging to detect in Arabic due to the language's complex structure. Methodology: The ADPDM model and framework for detecting plagiarism in Arabic documents are presented in this dissertation. It is designed to detect plagiarism within academic contexts. By organizing documents logically into paragraphs, sentences, and words, the model seeks to establish a robust system that can identify duplicated content and search for similar documents within identifying corresponding sets. In particular, the study examines preprocessing techniques such as stop word removal, stemming, and rootage processing, followed by content-based methods utilizing fingerprinting and heuristic algorithms that are tailored to Arabic language features. To aid in efficient detection, the BKDR hash function is used for chunk having. To optimize computation time, heuristic algorithms are implemented at different levels of document representation, using metrics such as Longest Common Substring (LCS). To evaluate the ADPDM system, a corpus of 100 documents is utilized, which includes datasets from AraPlagDet and the Decision Support System (DSS). The performance of ADPDM is compared with other plagiarism detection methods using WCopyFind, but the latter has a higher computational speed than ADBDM. Its recall, precision, and F-measure values of 0. 78035, 0. 994264, and 0. 865688 respectively (ADPDM) are particularly notable for its ability to detect plagiarized content in Arabic documents. ADPDM is a successful anti-pluralist solution for Arabic text, even though it requires a longer processing time than WCopyFind.

Short-Term Energy Generation Forecasts at a Wind Farm—A Multi-Variant Comparison of the Effectiveness and Performance of Various Gradient-Boosted Decision Tree Models

High-quality short-term forecasts of wind farm generation are crucial for the dynamically developing renewable energy generation sector. This article addresses the selection of appropriate gradient-boosted decision tree models (GBDT) for forecasting wind farm energy generation with a 10-min time horizon. In most forecasting studies, authors utilize a single gradient-boosted decision tree model and compare its performance with other machine learning (ML) techniques and sometimes with a naive baseline model. This paper proposes a comprehensive comparison of all gradient-boosted decision tree models (GBDTs, eXtreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), and Categorical Boosting (CatBoost)) used for forecasting. The objective is to evaluate each model in terms of forecasting accuracy for wind farm energy generation (forecasting error) and computational time during model training. Computational time is a critical factor due to the necessity of testing numerous models with varying hyperparameters to identify the optimal settings that minimize forecasting error. Forecast quality using default hyperparameters is used here as a reference. The research also seeks to determine the most effective sets of input variables for the predictive models. The article concludes with findings and recommendations regarding the preferred GBDT models. Among the four tested models, the oldest GBDT model demonstrated a significantly longer training time, which should be considered a major drawback of this implementation of gradient-boosted decision trees. In terms of model quality testing, the lowest nRMSE error was achieved by the oldest model—GBDT in its tuned version (with the best hyperparameter values obtained from exploring 40,000 combinations).

Advances and Challenges of Single‐Pixel Imaging Based on Deep Learning

AbstractSingle‐pixel imaging technology can capture images at wavelengths outside the reach of conventional focal plane array detectors. However, the limited image quality and lengthy computational times for iterative reconstruction still hinder its practical application. Recently, single‐pixel imaging based on deep learning has attracted a lot of attention due to its exceptional reconstruction quality and fast reconstruction speed. In this review, an overview of the current status, and the latest advancements of deep learning technologies in the field of single‐pixel imaging are provided. Initially, the fundamental principles of single‐pixel imaging and deep learning, followed by a discussion of their integration and associated benefits are presented. Subsequently, a comprehensive review is conducted on the advancements of deep learning in various domains of single‐pixel imaging, covering super‐resolution single‐pixel imaging, single‐pixel imaging through scattering media, photon‐level single‐pixel imaging, optical encryption based on single‐pixel imaging, color single‐pixel imaging, and image‐free sensing. Finally, open challenges and potential solutions are discussed.

Stochastic Vehicle Routing With Delivery Choice

ABSTRACTWe consider the problem of designing delivery routes for vehicles where the vendor has the choice of how much of the demand from a customer to fulfill. The customer demand is known a priori only as a probability distribution. Exact customer demand is known only after visiting the customer. Different customers are able to negotiate different prices for each unit of product with the vendor. Given a route, the objective is to decide at each customer location, how much demand to satisfy so as to maximize expected profit taking into account a linear penalty cost for unfulfilled demand and the vehicle routing costs. In this article, we develop several new structural results for this problem. We illustrate how these structural results can be embedded in different heuristic frameworks commonly used for deterministic vehicle routing problems. This helps develop efficient routes for a single vehicle as well as a multiple vehicle scenario for this stochastic variant. For small‐sized problems that allow for exhaustive enumeration, we demonstrate the effectiveness of the illustrated heuristic. For larger problem instances, based on structural results, we develop methods that allow the heuristic to run more efficiently than otherwise. Results are reported on instances based on benchmark instances drawn from literature for upward of 100 customers and vehicle capacity up to 600 units. Computational times needed to heuristically solve such problems are within 1 100 s.