You are here

AI/ML for Nitramine Recrystallization and Coating


OUSD (R&E) MODERNIZATION PRIORITY: Artificial Intelligence/Machine Learning TECHNOLOGY AREA(S): Information Systems TOPIC OBJECTIVE:  To develop a suite of probe technology and machine learning algorithms which can be used throughout the energetics manufacturing process to reduce cost and increase product consistency. TOPIC DESCRIPTION:  Currently, nitramine energetic materials have unacceptably high rework/scrap rates in a number of different munitions’ energetics manufacturing processes, such as dissolution, recrystallization, and slurry coating. This is largely due to the plant operators inability to control critical manufacturing parameters such as cooling water temperature, nitramine concentration, and solvent/antisolvent ratios. To further exacerbate the problem, munitions’ energetics manufacturing processes are poorly understood ‘black boxes,’ so the reason behind any deviation from spec is difficult to ascertain. We believe that by deploying a number of different measurement probes during various manufacturing steps, and analyzing the data via machine learning, we can dramatically reduce or even eliminate out of spec batches. The probes will collect data in real time as materials are manufactured, and the machine learning algorithms will provide nearly instantaneous recommendations to the plant operators on how to adjust their processes to target the desired properties. Beyond reducing out of spec batches, we would also like to reduce cost, environmental footprint (by mainly increasing energy efficiency and reducing solvent use), and increase throughput from existing lines. We believe in the long run, the insights gained for this program will enable easier transition of novel energetic formulations.   The purpose of this topic is to explore probe technology and machine learning algorithms that will be examined from the offerors and will be down selected based on time resolution, ruggedness, data output, safety, and suitability. There are two key steps for this technology to be successful. The first is the development and deployment of novel probes that produce large amounts of data in real time. The second is an advanced machine learning system that can take the probe readings and provide adjustments in real time during manufacturing to produce the desired product. PHASE I: Demonstrate proposed probe technology can produce the data required over the course of several manufacturing runs. The Phase I Base amount must not exceed $100,000 for a 6-month period of performance. PHASE II:  Implement the probes during the manufacturing process and collect data. Use data with machine learning algorithms. • Phase II Sequential: Expand the probes to more manufacturing lines, increasing the amount of data for machine learning systems. Direct and control energetics manufacturing based on machine learning recommendations to realize benefits PHASE III and DUAL USE APPLICATIONS:  Applying AI/ML to chemical manufacturing has a ton of commercial potential, although this specific topic is geared towards energetics; therefore, landing this topic at moderate dual-use potential for commercial capabilities. All HMX/RDX batches are currently examined via a multitude of techniques to determine if they are meet specifications. These include assessing purity, particle size, and thermal stability. This analysis can compared against the predictions of the machine learning algorithms and the measurements provided by the probes. We will also test the machine learning predictions against the predictions of crystallization modeling software, when appropriate.   KEYWORDS:    Probe; Algorithms; Machine Learning; Manufacturing process; Energetic materials; Energy   REFERENCES:  1. 2. 3.
US Flag An Official Website of the United States Government