Damage control remains a vital task aboard US Navy vessels, requiring highly accurate and rapid detection of hazardous events and a significant commitment of manpower. Traditionally, crew members monitor compartments and provide assessments verbally to damage control personnel and the ship's command. However, manual remote monitoring, (e.g., by video surveillance) of sensor data by human observers leads to rapid attention fatigue and, consequently, to missed detections.
The need for automated monitoring of spaces aboard ship and assessment of events such as chemical dispersal, toxic spills, and fire or flood detection, has stimulated development of multisensor, multicriteria sensing systems.1 Theoretically, multimodal sensing platforms offer benefits over more conventional point detection systems in terms of robustness, sensitivity, selectivity, and applicability.2,–5
Automatic, real-time monitoring of ship compartments with low-cost sensors connected to the existing ship infrastructure, including video surveillance, may provide significant reductions in damage control costs. This should reduce manpower requirements while offering the potential for enhanced detection capabilities. Such networked systems present a unique set of challenges.1,6
The US Naval Research Laboratory (NRL) has developed Volume Sensor (VS), an automatic monitoring system for standoff identification of damage control events, including fires, explosions, pipe ruptures, flooding, and gas release. A component of the Advanced Damage Countermeasures part of the Office of Naval Research's Future Naval Capabilities program, VS can reduce false positives and improve response times.
Figure 1. Example of a sensor suite made with off-the-shelf components.
The VS was built from low-cost commercial-off-the-shelf (COTS) hardware components integrated with intelligent software and smart data fusion algorithms developed at NRL. Sensors are organized into `suites’ as shown in Figure 1. Each suite contains a video camera, a long-wavelength (near-infrared) filtered camera, two single-element spectral sensors (UV and IR), and a human-audible microphone. Sensor integration is achieved with a modular system architecture as charted in Figure 2. An extensible, hierarchical data structure enables fast access and rapid transfer of sensor data.
Figure 2. Volume Sensor system architecture.
Event recognition is achieved in real-time with a tiered data analysis cycle. Data are acquired and analyzed by sensor-specific algorithms, then passed on to a fusion machine in which data from multiple sensors are analyzed by event-specific data fusion algorithms. Events flagged in the current cycle are compared with events from prior cycles. Event persistence is evaluated and threat levels forwarded to a supervising damage control system. A Bayesian approach is taken at each step of the analysis, with parameters estimated from the experimental testing of sensors and simulation of the algorithms.
Figure 3. Correct classifications of fire sources and rates of false positives due to nuisance sources for Volume Sensor Prototypes (VSP), video image detection (VID), and spot-type (EST) detectors.
Two Volume Sensor prototypes7 (VSPs) were constructed and fully evaluated aboard the ex-USS Shadwell in competition with two video image detection (VID)8,9 and three spot-type10 detection systems by Edwards Systems Technology (EST). A variety of incipient damage control and nuisance sources were employed to test early warning capabilities of each system.
The VSPs successfully monitored a variety of compartments, detected and discriminated multiple simultaneous and consecutive damage control and nuisance events, and conveyed timely situational awareness to a supervisory control system. For fire sources, results are summarized in Figure 3. The VSPs were much faster than VIDs in detecting flaming fires, due to the rapid response of the long wavelength camera and spectral sensing components, and to the ability of these sensors to perceive fire radiation reflected from walls and obstructions. The VSPs were initially slower than the VID systems at detecting smoldering fires, but achieved comparable detection rates within 30 seconds of the first alarm. The VID systems performed quite well at detecting fire sources, but markedly underperformed at nuisance rejection. The EST point detectors were generally slower and less consistent across event classes than the VSP and VID systems. The VSPs also provided effective situational awareness of pipe ruptures, flooding scenarios, fire suppression system activations, and, in addition, gas release events, which was not possible with the commercial systems.
Full-scale testing and evaluation of Volume Sensor prototypes—in competition with standard and state-of-the-art commercial fire detection systems aboard the ex-USS Shadwell—confirmed the utility of a multicriteria, multisensor approach to real-time damage control. Better overall performance was achieved using smart data fusion algorithms to combine and analyze real-time data from multiple, distributed sensors. Further, the Volume Sensor approach provides a useful template for the integration of heterogeneous sensors into networks for a variety of real-time sensing and situational awareness applications.
Christian Minor, Daniel Steinhurst
Nova Research, Inc.
Dr Christian Minor is a senior staff scientist at Nova Research, Inc. and performs research under contract in the Chemometrics and Chemical Sensing section at the US Naval Research Laboratory. He received his PhD in astrophysics from the University of California at Riverside in 2004. Dr Minor has been instrumental in coordinating, formulating, and implementing data collection and analysis algorithms and data-fusion methodologies necessary for distributed and multisensing applications.
Susan Rose-Pehrsson, Kevin Johnson, Jeffrey Owrutsky, Stephen Wales
US Naval Research Laboratory
Hughes Associates, Inc.