TY - JOUR
AU1 - PhD, W. Bradley Fain,
AU2 - FAAFP, Shean Phelps, MD, MPH,
AU3 - PhD, Alessio Medda,
AB - ABSTRACT In recent U.S. military experience, widespread exposure to improvised explosive devices has been implicated in noticeable changes in the incidence of brain injuries inversely related to reduced mortality—thought to be the unintended consequence of increase in exposure to blast wave effects—secondary to improved vital organ protection, improved personal protective equipment. Subsequently, there is a growing need for the development and fielding of fully integrated sensor systems capable of both capturing dynamic effects (i.e., “blast”) on the battlefield—providing critical information for researchers, while providing value to the medical community and leaders—for development of pre-emptive measures and policies. Obtaining accurate and useful data remains a significant challenge with a need for sensors which feed systems that provide accurate interpretation of dynamic events and lend to an enhanced understanding of their significance to the individual. This article describes lessons learned from a data analysis perspective of a collaborative effort led by a team formed at Georgia Tech Research Institute to develop a “sensor agnostic” system that demonstrates full integration across variant platforms/systems. The system is designed to allow digital and analog time/frequency data synchronization and analysis, which facilitated the development of complex multimodal modeling/algorithms. INTRODUCTION In recent U.S. military experience, widespread exposure to improvised explosive devices (IEDs) has been implicated in noticeable changes in the incidence of brain injuries. Whereas reduction in overall service member mortality—likely through improved vital organ protection—is attributed to improved personal protective equipment (PPE), evolving threats such as enemy employment of IEDs and insurgent tactics are widely considered to increase risk of significant “blast wave” exposure. This evolving “threat versus protection” paradigm shift is suspected of producing the unintended consequence of resulting in an overall increase in neurologically significant blast wave exposures in the modern warfighter. With the emergence of recent conflicts in Iraq and Afghanistan, the number of explosion-related injuries for U.S. service members has increased considerably. IEDs were responsible for 40% to 60% of service member injuries/deaths on the battlefield in Iraq between 2006 and the summer of 2009. During that same period, IEDs were responsible for 50% to 75% of service member fatalities incurred in Afghanistan.1 Direct exposure to blast from IEDs while on foot can result in penetrating injuries, whereas serious injuries are reported in mounted scenarios involving U.S. combat vehicles exposed to roadside IEDs from underbody blasts. As such, brain injuries are among the most commonly reported injuries amongst military personnel, and traumatic brain injury (TBI) has been termed the “signature wound” of the current conflict. It is estimated that between 2001 and 2007, about 63% of all TBI cases were associated with an explosive event, and about 15% to 25% of all service members deployed in Iraq reported TBI.2 There exists a fundamental need to better understand and quantify the direct and indirect effects of blast to individuals engaged in modern warfare. The last 10 years have witnessed an expansion of effort to understand explosive blast effects on vehicles and personnel as well as to develop systems targeted for testing and deployment to the field. Some of these efforts are focused on platform systems but the majority are geared toward service members and designed to capture the effect of a blast environment on the human body and the brain. The purpose of collecting data is to determine how and why these injuries are occurring and to devise methods to quantify them, and, thereby, better protect against, or even prevent, deleterious outcomes. Obtaining accurate and useful data in theater is and remains a significant challenge. Theater data collection must be well and fully executed to enable accurate description of operational exposures and development of adequate preventive measures. As such, there exists a pressing need to develop and field fully integrated sensor systems that more accurately record, store, and forward data that accurately describes salient physical and environmental phenomena. Moreover, such sensors must feed extant and burgeoning systems that provide accurate interpretation (in real time as well as postevent forensic analysis) of these dynamic events and lend to an enhanced understanding of their significance to the individual and to the fighting force as a whole. Accordingly, in the period between 2007 and 2011, limitations encountered in the accurate collection and analysis of dynamic laboratory, field, and postevent blast event data of deployed (and/or emerging) systems highlighted the need for a more “unified” approach to operational data collection and investigation. Focused high-level discussions between the Vice Chief of Staff of the U.S. Army and leading scientists/technologists resulted in a directive ordering the assessment of technologic readiness and/or feasibility of the development of an integrated (meaning a system that tied vehicle or “platform” data to individual Warfighter data), government-owned, “sensor agnostic—system of systems” that would allow accurate and reliable time–place–frequency spectrum synchronization and individual localization/proximity detection regards an injury causing dynamic blast event. The Integrated Blast Effects Sensor System (IBESS) team was formed at Georgia Tech Research Institute (GTRI) and evolved a “sensor agnostic” system that would (1) demonstrate full integration across variant platforms/systems, (2) allow digital and analog time/frequency data synchronization and analysis, (3) complex multi-modal modeling/algorithm development, (4) provide synchronization of data across both Warfighter and vehicle platform systems (i.e., when an event occurred while mounted), and (5) offer the possibility of an expandable “fusion platform” from which future sensor systems with diverse, multiple streams of time point data could be collected, assessed and analyzed for dynamic operational events.3 Furthermore, such a system was envisioned to provide new insights into other, not well understood significant operational exposures, such as rollovers, collisions, and/or kinetic vehicle strikes. This article describes an overview of a parallel effort that coalesced many lessons learned in the process of developing, testing, and fielding IBESS—a fully integrated system of both individual and vehicle platform sensors—to record explosive event signatures. Important considerations are included (see Medda A, et al: Final Report on Lesson Learned: Implementation of a data acquisition system for the recording of explosive events signature, Georgia Tech Research Institute, Atlanta, GA 2013. SENSIAC Technical Report, OSP D6556) for data collection, processing, interpretation (see Medda A, and Fain B: Blast Data Analysis Final Report. Georgia Tech Research Institute, Atlanta, GA 2013, SENSIAC Technical Report, OSP D6556), and archiving. The purpose of this report is to provide a resource for evaluating current systems and offer a set of guidelines for designing new systems. The work described was a collaborative effort between GTRI as government's prime contractor, with L-3 Applied Technologies, Inc. in a supporting subcontractor role, in direct support of the U.S. Army Rapid Equipping Force's efforts to tackle the multiple issues revolving around accurate assessment and quantification of dynamic blast event data in the combat environment (see Rapo M, et al: IBESS Applique Analysis, Algorithms, and Techniques Final Report. L-3 Applied Technologies Inc., San Diego 2013, Technical Report, J0594-13-568). LESSONS LEARNED One of the key factors to the projects' success was the fact that data collection must begin and terminate with the desired outcome in mind, that is, correlation to individual injury data. Correlation of sensor data with medical data requires a model with an accurate biomechanical representation of the injury/ies of interest. The model requires accurate data that can be measured and/or calculated from the blast event to provide a validated estimation of the resulting effect on the human body. Blast event sensors must be strong enough to withstand environmental conditions yet small, lightweight and integratable into current PPE to be worn effectively by the service member. Sensors must be cost-effective and mass-producible yet accurate enough to capture the dynamic range and frequency content of the blast event and associated outcomes. Sensor correction algorithms must account for onboard amplifying and filtering effects, as well as nonideal mounting, location, and orientation effects. The theater is the harshest environment for sensors, and data anomalies must be identified and corrected. Sensors must be correctly associated with their service members and the correct location that the sensor is placed on the service member's body. The time stamp needs to be accurate enough to associate events between sensors and in the medical records. Accordingly, data retrieval, processing, and storage must follow a standard procedure for all Field Service Representatives (FSRs). FSRs provide critical in-field technical support to the unit and users while acting as the key communications link between the unit and the technology/engineering teams involved so as to maximize both quality and quantity of data retrieved and passed on for analysis. Fundamental properties of the blast event and the resulting effects to the warfighter also need to be considered. The number and type of sensors selected will depend on meeting requirements of the measurement range, as well as how they perform based on mounting position and orientation on the PPE. Additionally, there are power, memory, and data quality constraints. Deciding between independent or integrated system architectures will determine communication protocols, power distribution, and data capture and transmission strategies. The package must be ruggedized and subjected to environmental and other types of laboratory and field testing. A protocol for collecting, preserving, and analyzing the data must be developed before deployment. Of utmost importance is the proper association of each service member with their respective system of sensors, along with accurate time-stamping of recorded events so that injuries documented in medical records can be properly correlated to measured and derived quantities. Once a dose injury curve has been established, the thresholds that separate the severity level (green, amber, and red) need to be set with guidance from the medical community. What needs to be measured will depend on each particular injury mechanism. There must be an injury model that links some external property or physical observable event to the injury mechanism. For example, the time history of the external pressure to the body is provided as input to the U.S. Army's lung INJURY 8.3 model, which calculates the work done on the body and provides a probability of lung injury. If circumstances precluded measuring that particular quantity, such as interference from protective gear, then another quantity could be measured as long as there was a way to then predict or derive that desired quantity. When the mechanism of injury is unknown, or an injury model has not yet been established, then a determination of the measurable properties needs to be made. These are the physical properties which characterize the blast event, such as the time history of pressure, temperature, density, air and particle velocities, and derived quantities such as impulse. There is the time history of the kinematics of the body, such as changes in posture and orientation through linear and angular motions, and the kinematics and dynamics of the interactions with the PPE. Physiological effects, such as increase in heart rate or core temperature, may need to be indirectly measured. Sensor Selection Choosing adequate sensor technology will depend on what type of data needs to be obtained. Piezoelectric (measurement of changes in pressure, acceleration, strain or force by converting them to an electrical charge) and piezoresistive (measurement of changes in the electrical resistivity of a semiconductor or metal wherein a mechanical strain is applied) transducers traditionally have been used to measure blast overpressure, whereas pyroelectric (measurement of the change in temporary voltage when heated or cooled) transducers measure heat. Tri-axial accelerometers (devices that measure proper acceleration—aka: “g-force”—along the three principal axis: x, y, and z) have been used to measure the motion of whatever the sensors are affixed to. Blast waves produce rapid changes in pressure, heat, and acceleration resulting in widely variable measures of strain and stress on both biologic and nonbiologic materials. Accordingly, the properties of an explosive event require the ability to accurately measure a wide range of magnitudes and frequency content of a given blast event. The blast wave begins with a jump to its maximum value, decays past zero within approximately 1 to 6 milliseconds into a negative phase, and then gradually recovers to zero. A sampling rate of 50 to 100 kHz will provide 50 to 100 data points within a 1-millisecond time period. Sensors will be limited in their response to the full dynamic range because of reliance on power supplied by a battery. Signal amplification and other onboard electronics often introduce a low-pass filtering, further reducing the dynamic range. At the very least, the sensor should be able to reliably measure the conditions associated with survivable events. Sensor memory must be large enough to store up to 100 milliseconds of data per event for multiple events. Commercially available products can help reduce cost as long as they can meet the data quality requirements. Sensor drift is common, but baseline changes in bias from shock impact must be considered. In addition to cost per unit, human factor requirements of size and weight will have a significant impact on the number and size of components that can be included in a complete sensor package. Battery life is limited by battery size and the demands placed upon it to keep some components in a state of readiness. System Architecture, System Integration, and Prototyping The number of sensors needed, and their specific placement on the body or PPE, will depend on modeling requirements. In general, a single independent sensor is not able to adequately characterize a blast event. Pressure sensors are directionally sensitive, measuring 2 to 8+ times the incident level when hit directly (“face-on”), while diminishing to incident levels when grazed (“side-on”). In regions that are “shadowed” or blocked by the body, the pressure values are less than incident level. We would ideally prefer to directly measure the forces experienced by the body part of interest such as acceleration and deceleration of the brain. However, practical limitations require us to place measurement devices externally to the body and often on top of body armor. Extensive modeling is required to account for body–pad–armor system interactions and the distance from the sensor to the actual body part of interest. Consideration needs to be given between having independent or integrated sensor architectures. It is very challenging (or near impossible) to time-sync multiple independent sensors in a way to meaningfully use arrival time differences. Because of clock drift, which can be minutes to days to weeks in error, it can be difficult to even align recorded events between sensors. Having the ability to place each sensor independently can make it adaptable to all kinds of PPE variations, but it could get placed outside the range of documented use required by the models. Wired integrated sensor architectures solve the placement and time-syncing issues, but may be difficult to integrate with the PPE. Wireless capabilities allow for independent placement and the ability to align recorded events between sensors, but communication time is not fast enough to capture blast wave arrival time differences. Communication protocols and data capture strategies need to be established for both the particular model dependent data acquisition system and for integration with other sensor systems. This information must be made available by the vendors in order for this to occur. One way to facilitate cooperation is to have a centralized hub, such as on a vehicle, which collects information from all sensor systems and is in frequent contact with the warfighter. With essentially unlimited power and processing capabilities, the system can provide automatic and continuous data collection, real-time analysis, and serve as a way to check the status of sensor health and update time-stamp information. Furthermore, experience demonstrates that providing such capabilities would reduce dependency on FSRs freeing these critical human resources to expend valuable time and effort elsewhere. Human Factors Human factors play an important role in the design and deployment of a field data acquisition system. A primary goal guiding the design of a sensor system should be that it is “operationally transparent,” minimizing to the extent possible any influence it might have on the conduct of a mission. Discomfort and interference with normal use of gear and critical safety equipment should be minimized. The full range of personnel sizes and types of gear should be accommodated in resultant designs not only for usability/wearability, but also to allow for full inclusion in the algorithmic (i.e., anthropometric consideration) interpretation of event data. In designing for the future, developers should consider additional safeguards to reduce the occurrence of equipment tampering. Feedback to maintainers should be examined, including the form, frequency, and method of delivery. Wherever possible, procedures should be simplified and information should be provided to maintainers at a level sufficient to support any decision-making required for successful completion of assigned tasks. With a weight often in excess of 75 lb, standard loads carried by the warfighter already present a substantial physical burden and the potential of a significant operational burden or safety risk. The system must not interfere with personal protective gear or normal service member activities. Ideally, the system should not require the service member to interact with the system under normal operations. Impact of Environmental Conditions The sensors and/or hubs must be able to withstand severe impacts and accelerations over multiple events. The components must not come unsoldered, broken, or severed, and external wires need protection at joints and between sensors. Potting sensors with epoxy can help stabilize and protect electrical and sensing components. Additionally, it can act as a sealant to protect against water and moisture. Housing materials and sensing components need to be resilient to extreme variations in temperature (−20 to 140°F), and to block interference from electromagnetic pulses. Sensors that have openings need meshes or other specialized coverings to protect against projectile particulates. Sensor performance needs to be evaluated as these screens become progressively blocked through the accumulation of dust, sand, or mud. Sensors must be characterized for actual performance to blast through shock tube and field testing. The shape of the sensor housing can introduce its own artifacts to the localized pressure field as the blast wave passes. The body and PPE can block or “shield” the sensor from the blast in some directions, or see multiple reflections from nearby body parts or equipment, causing the sensor to behave differently than when hit unimpeded. Nonrigid mounting effects from clothing and other PPEs, as well as blunt impacts from projectiles, can completely change the signal, and thus must be classified and corrected. Testing and Evaluation Testing is an essential component in the design of a ruggedized sensor suite. Performance of the product can be evaluated from concept to completion. Each component of the sensor system must be tested for durability and functionality under different conditions, for instance, blast or blunt impact. The system processes themselves can be tested, such as the acquisition and storage of data. Mounting locations and sensor schemes can be tested using surrogate human dummies. Any number of test methodologies can be developed to investigate the number of variables or issues that arise during the design process. For the sensor systems discussed in this report, shock tube testing will be a vital tool in assessing performance under blast conditions. Shock tube testing is an economical and quick method to obtain blast wave sensor responses and weed out artifacts in the data resulting from the system design. After extensive shock tube testing has been conducted, field testing using real explosives will provide a more powerful yet less economical method of testing the sensor system. Accurate combat data can only be obtained if the equipment has been fully characterized and all issues seen in the testing process are resolved. Fielding Sensors must be correctly associated with a particular service member and its location on the body. The device ID should be computer generated, the service member ID should be obtained through the scanning of a Common Access Card or similar card (not entered manually), and this information should appear in the header-file of the device. The sensor must either determine its location on the body relative to other sensors in the system, or it needs to have an external label that says where the sensor is to be placed and oriented. Sensors from multiple service members can end up in piles together, so the identification information must be accurate and retrievable. The event file must be fully self-documenting. Having someone else (like an FSR or analyst) fill data fields often leads to irresolvable inconsistencies. Raw event files must be preserved, with data extracted and stored in a database at a later time. Data collection must be standardized, with FSR's following the same standard operating procedure, including how files are transferred (via ftp, mail, etc). A consistent method of naming the data files needs to be established from the beginning. Deployment Sensor system deployment is a large part of a successful sensor system as any perceived issues with deploying the system will reflect directly to the product's image and ultimate success. Effective deployment of sensors should streamline accountability and identification of the sensors, both of which ultimately facilitate the collection of data at later dates. Data Analysis Most of the expected sensor responses will be identified through laboratory and field testing. Theater exposure always introduces unknown signal artifacts that must be detected, classified, and corrected. Representative samples of data must be collected and analyzed, including both “good” and “bad” data. Nonevent data is useful during the development of screens and filters. Once sensor signals have been corrected for bias (difference between an expected value and the true value of the parameter being estimated), drift (changes in statistical properties over time in unforeseen ways), and noise (recognized amounts of unexplained variation in a sample), the data sample should be downsized for conservative possible “real” events. The sample can be further reduced by removing physically impossible events. Signal metrics can be used to screen for “interesting” and “plausible” events, and flagged for review by an analyst. This can include classification of other known events, such as electromagnetic interference, small arms fire, nonblast shock and impacts, etc. It can also include changes to the blast wave caused by the presence of walls, ground, vehicles, other service members, or from being inside an enclosed area (room or vehicle). The techniques developed for screening will also depend on the type of data being recorded (e.g., acceleration vs. pressure), the quality of the data, sensor specific corrections formulated during the testing phase, and predictive modeling capabilities (e.g., biophysical modeling, signal processing techniques). Medical Data Correlation The objective is to verify the dose–response model and thresholds. Data correlation (i.e., assignment of statistically relevant dependency between observed variables) to medical data should be guided by injury biomechanics. Only usable data established by theater screening algorithms should be used for medical data correlation. The dose–response models proposed by the system developer need to be tested against theater data to set injury thresholds. The theater data set should consist of as much usable data as possible. The inclusion of no-injury data is important for determining the injury threshold. Multiple candidates for the dose function, such as peak pressure, impulse, and including multivariate combinations of peak pressure and duration can be hypothesized for testing. Statistical analysis will be performed to generate the correlations for each proposed dose function. The correlations will be ranked by appropriate goodness-of-fit tests to identify the best fit. Confidence bands should be calculated to evaluate the data fit and identify outliers for additional analysis. Appropriate threshold settings for assessing and denoting level of exposure should be developed in close consultation with both medical and operational commands. CONCLUSION The successful design and fielding of a sensor system requires that the technical and logistic planning phases occur with foresight of pitfalls. The purpose of this article was to share a number of critical lessons learned that will enable the next generation of fielded sensor systems to lay a foundation that will lead to its eventual success. The end purpose must be kept in mind from the beginning. In many cases, the ultimate goal is to determine how injuries are occurring to our warfighters so that mitigation and prevention strategies can be developed. The following critical lessons learned directly must be addressed in any future successful blast research program: — The data collected must be correctly associated with the events and injuries recorded. Poor record keeping and missing data fields can render otherwise excellent data useless. Clock drift can lead to inaccurate time stamps, potentially thwarting data association. — From a logistics perspective, a support structure must be in place for the real-time retrieval of the data. — Sensor systems must be worn as designed, or interpretation of the data becomes almost impossible. — Data must make it back to the organizations and people who analyze it, and in its raw form. It cannot be manipulated or held up as it passes hands. Wherever the mechanism of injury is known and established injury models exist, the goal is to acquire input data to validate (i.e., statistically corroborate known exposures and effects) methodologies to collect relevant and actionable inputs to a model for further analysis and improvement. When requisite phenomenological data cannot be directly measured, transfer functions must be developed to “approximate” associations and correlations. The assumptions and limitations of such transfer functions must be fully disclosed so that the analyst can determine when event conditions and data fall outside of their validated ranges of operation. A systems perspective must be provided for the measurements made by each sensor. Sensors measure local effects only. Nonrigid mounting conditions, local surface geometry, nonstandard features (e.g., interfering loadout gear), and deviances in mounting location and orientation can all significantly affect what the sensor measures. Additionally, changes in the initial location and posture of the warfighter relative to the originating event will result in sensors measuring very different quantities. Multisensor algorithms are required to provide interpretation of the data. Sensor systems will require correction algorithms. The particular sensing elements chosen will likely not perform to the level of commercial grade sensors. Furthermore, onboard amplifiers and data acquisition solutions will degrade the overall performance. What is finally recorded will not be identical to what actually occurred, even in laboratory environments. Laboratory and field testing of the sensor components and sensor systems as a unit are exceedingly valuable for characterization of performance under known and controlled conditions. As problems arise, rapid feedback and design iteration can occur. In preparation for fielding, stressing conditions can be tested (environmental, projectile and blunt impacts, etc.). Field and theater environments are harsher than laboratory tests. Screening algorithms will be required to remove signal artifacts and eliminate erroneous events. The lessons learned documented in this article may help explain some of the reasons why the community has not yet achieved the goal of establishing a valid correlation between physical injury and blast dosage. Research should continue on both the sensor systems and blast data analysis fronts. Furthermore, a better integration between the engineers developing sensor systems and the scientists responsible for data analysis would likely result in a better overall system solution. ACKNOWLEDGMENTS A portion of the work described in this article was funded by the Military Sensing Information Analysis Center (SENSIAC) under delivery order 187. SENSIAC is sponsored, administratively managed, and partly funded by the Defense Technical Information Center-Administrative Instruction under contract HC1047-05-D-4000. The SENSIAC blast data analysis project was a collaborative effort between GTRI as the prime contractor and L-3 Applied Technologies, Inc. (L-3/ATI) as the subcontractor in support of the U.S. Army Rapid Equipping Force. REFERENCES 1. Lenhart M, Savitsky E, Eastridge B Combat Casualty Care: Lessons Learned from OEF and OIF . Borden Institute, November 9, 2012. Available at http://www.cs.amedd.army.mil/borden/book/ccc/UCLAFrontMatter.pdf; accessed November 8, 2013. 2. Wojcik B, Stein C, Bagg K, Humphrey R, Orosco J Traumatic brain injury hospitalizations of U.S. army soldiers deployed to Afghanistan and Iraq. Am J Prev Med  2010; 38( 1 Suppl): S108– 16. Available at http://www.researchgate.net/publication/51441598_Traumatic_brain_injury_hospitalizations_of_U.S._army_soldiers_deployed_to_Afghanistan_and_Iraq; accessed November 8, 2013. Google Scholar CrossRef Search ADS PubMed  3. Mulkey N, Liu B, Medda A The Integrated Blast Effects Sensor Suite: A Rapidly Developed, Complex, System of Systems. Proceedings of the 8th International Conference on System of Systems Engineering (SoSE) , 2013; 224– 8. Available at http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6575271; accessed June 29, 2014. Footnotes 1 This article was presented at the Military Health System Research Symposium, Fort Lauderdale, Florida, August 12–15, 2013. Reprint & Copyright © Association of Military Surgeons of the U.S.
TI - Lessons Learned From the Analysis of Soldier Collected Blast Data
JF - Military Medicine
DO - 10.7205/MILMED-D-14-00431
DA - 2015-03-01
UR - https://www.deepdyve.com/lp/oxford-university-press/lessons-learned-from-the-analysis-of-soldier-collected-blast-data-FM60D4090r
SP - 201
EP - 206
VL - 180
IS - suppl_3
DP - DeepDyve
ER -