Neural network for quality control of submunitions produced by injection loading Page: 4 of 7
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
The following text was automatically extracted from the image on this page using optical character recognition software:
NEURAL NETWORK FOR QUALITY CONTROL OF SUBMUNITIONS
PRODUCED BY INJECTION LOADING
Ronald E. Smith', William J. Parkinson', Ralph F. Hinde, Jr. , Kirk E. Newman2, and Paul J. Wantuck'
Engineering Sciences and Applications Division'
Los Alamos National Laboratory
Los Alamos, NM 87545
Energetic Materials Research and Technology Department2
Naval Surface Warfare Center, Indian Head Division
Yorktown, VA 23691
Injection loading of submunitions for smart weapons is a novel automated processing technique that can benefit
from adaptive process control. This paper describes how the quality of submunitions could be controlled by using a
neural network code in real time. Future work is planned to demonstrate fewer rejects and pollution reduction
during submunition manufacturing.
Neural Networks; Smart Weapons; Submunitions; Pattern Recognition; Process Control; Injection Loading
The injection loading device discussed in this paper  was developed in order to produce submunitions filled with
highly viscous material. Each run fills multiple submunitions serially, with high viscosity plastic bonded explosive
(PBX) material. Even though most loads are acceptable and meet specifications, the rejection rate is often excessive.
The loads that are not acceptable present a severe waste disposal problem. The current technology requires an X-ray
inspection after the submunitions have been loaded with PBX. Loads that have voids, or that have density gradients
will be rejected. Since operating conditions leading to rejectable products may occur at the beginning of an injection
cycle, aliquots of ten pounds of PBX and the corresponding submunitions loaded may also be rejected. Therefore, it
is imperative that the process control algorithm recognize disturbances in the injection loading parameters as soon
as possible so that corrective action can be taken to resolve the upset before the PBX loading is completed. Each
load requires about 30 seconds. In an effort to determine submunition quality on-line, the device was instrumented.
Ram displacement, cavity pressure, and hydraulic pressure were measured at two second intervals for each load.
Ram velocity, shear rate, shear stress, and viscosity were calculated at each time interval. The data were then
compared with the post-mortem X-ray results to determine if the load was a pass or a fail. To the naked eye, no
patterns emerged in the injection loading data that correlated specific processing parameters to the submunition
pass/ fail criteria.
Several neural network models were applied before we found one that worked. A backpropagation model using two
inputs, cavity pressure and viscosity, and three hidden nodes finally cracked the code. This model, even with a
limited number of training sets, was able to predict bad or good loads 100%, after only four sensings, eight seconds
Here’s what’s next.
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Article.
Smith, R.E.; Parkinson, W.J.; Hinde, R.F. Jr.; Wantuck, P.J. & Newman, K.E. Neural network for quality control of submunitions produced by injection loading, article, December 1, 1998; New Mexico. (digital.library.unt.edu/ark:/67531/metadc675166/m1/4/: accessed October 15, 2018), University of North Texas Libraries, Digital Library, digital.library.unt.edu; crediting UNT Libraries Government Documents Department.