ML-based Prediction of Explosion Shockwaves

In a paper published in the journal Defence Technology, researchers developed two physics-informed machine learning (PIML) models to improve the accuracy of predicting peak overpressure of ground-reflected explosion shockwaves. These models substantially reduced prediction errors in datasets used for training and non-training. They also incorporated a physical loss term, enhancing the models' extrapolation performance. This study offered insightful information for structural design and explosion hazard assessment.

Study: ML-based Prediction of Explosion Shockwaves. Image Credit: Hussain Warraich/Shutterstock
Study: ML-based Prediction of Explosion Shockwaves. Image Credit: Hussain Warraich/Shutterstock

Related Work

Prior studies have shown that explosives that burst in the air produce destructive blast waves, essential for comprehending the explosives business and structural safety. Peak overpressure forecasting techniques such as mathematical frameworks, numerical simulations, and experimental measurements have inherent limits. Thanks to developments in computer technology, ML algorithms that use physical data have demonstrated promise in increasing prediction accuracy.

Predicting Ground-Reflected Overpressure

Extensive experimental research has established several models for predicting ground-reflected wave peak overpressure from trinitrotoluene (TNT) explosive detonations at ground level. These models, detailed in standards like GB6722-2003 and military specifications such as GJB 6930-2008, account for both regular and irregular (mach) reflections of shock waves.

They rely on parameters like scaled distance (Z), which relates the explosive-to-measurement point distance (R) to the explosive mass (M). Additional contributions include mathematical formulations by researchers to understand mach reflection initiation points and the transition between regular and mach reflections. These are crucial for accurate overpressure predictions on various ground surfaces.

Simultaneously, deep learning (DL) frameworks with integrated physical principles have given rise to PIML models, which aim to enhance prediction accuracy. Scaled blast height (Hs) and scaled distance (Rs) are two examples of input variables that PIML-1 uses to compute residuals that minimize disparities between physics-based computations and observed values. It uses a neural network structure optimized by 2000 iterations using the Adam method, consisting of three hidden layers with 64 neurons each.

PIML-2 builds on this approach and ensures that predictions closely match expected physical behaviors by incorporating regularization terms into the loss function that incorporates physical priors. Both models aim to improve peak overpressure predictions, which are essential for applications in structural safety and explosive danger assessment.

Dataset Creation Process

During the dataset creation process, several crucial measures must be taken to accurately and efficiently forecast ground-reflected wave peak overpressure from TNT explosive detonations. An autonomously dynamic network (AUTODYN) is used to develop a finite element model that uses two computational phases to simulate the expansion of the detonation and the propagation of shock waves in the air. Grid convergence calculations are carried out to optimize mesh sizes for both phases. The results show that a 1 mm grid size is appropriate for the first stage and a 10 mm grid size for the second.

With correlation values (R2) ranging from 0.872 to 0.995 and average relative errors ranging from 3.88% to 15.10%, model parameter validation offers more evidence for the model's ability to predict peak overpressure appropriately. Finally, a large-scale dataset encompassing different blast heights, horizontal distances, and TNT masses is developed. It is then standardized to improve the effectiveness of ML models before being used for training and testing. 

Integrated Explosive Event Prediction

The evaluation of physical models for predicting peak overpressure from explosive detonations provides valuable insights into their performance across different scenarios. Among the models assessed, the Hong Hao and GJB6930-2008 models stand out for their robust predictive accuracy, characterized by lower average relative errors and higher coefficients of determination (R2). These models accurately forecast peak overpressure under various blast conditions, demonstrating their effectiveness in capturing complex physical phenomena associated with explosive events.

Moreover, these physical models are more predictive when integrated into PIML frameworks, especially when extending outside training data. For instance, PIML-1 predicts peak overpressure more accurately in situations not seen during training since it blends the advantages of neural networks and physical models. This integration improves forecasts and accommodates a greater range of operational conditions.

Similarly, PIML-2 enhances predictive accuracy by incorporating a physical loss term into the ML model. This approach adjusts the model's weights to better adhere to physical constraints, improving its ability to generalize predictions beyond the training dataset. The inclusion of physical insights within ML enhances predictive performance. It ensures that predictions remain reliable and consistent across diverse operational environments, underscoring the synergy between physics-based modeling and advanced ML techniques in predicting explosive events.


To sum up, the study successfully integrated ML with physical model deviations to predict peak overpressure across diverse scenarios. Models like Ye Xiaohua and GJB6930-2008 were incorporated, significantly reducing average prediction errors from 17.459%–48.588% to 1.0822%–1.1434%. The models' improved predictive accuracy was demonstrated by the fact that over 99% of forecasts had relative errors under 20%.  

These models outperformed conventional physical models, achieving over 100% accuracy in forecasting peak overpressure values beyond the training data. Future research should optimize characteristics such as shockwave arrival timing and positive pressure duration. The study suggested a quick and efficient way to create blast shockwave power fields by employing physical-informed neural networks to make precise predictions.

Journal reference:
Silpaja Chandrasekar

Written by

Silpaja Chandrasekar

Dr. Silpaja Chandrasekar has a Ph.D. in Computer Science from Anna University, Chennai. Her research expertise lies in analyzing traffic parameters under challenging environmental conditions. Additionally, she has gained valuable exposure to diverse research areas, such as detection, tracking, classification, medical image analysis, cancer cell detection, chemistry, and Hamiltonian walks.


Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Chandrasekar, Silpaja. (2024, June 21). ML-based Prediction of Explosion Shockwaves. AZoAi. Retrieved on July 17, 2024 from

  • MLA

    Chandrasekar, Silpaja. "ML-based Prediction of Explosion Shockwaves". AZoAi. 17 July 2024. <>.

  • Chicago

    Chandrasekar, Silpaja. "ML-based Prediction of Explosion Shockwaves". AZoAi. (accessed July 17, 2024).

  • Harvard

    Chandrasekar, Silpaja. 2024. ML-based Prediction of Explosion Shockwaves. AZoAi, viewed 17 July 2024,


The opinions expressed here are the views of the writer and do not necessarily reflect the views and opinions of AZoAi.
Post a new comment

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.

You might also like...
Optimizing Wastewater Treatment with Machine Learning