False-high-rate alarm

The National Institute of Justice has put together multivolume compendiums of instrumentation relevant to chemical and biological weapons detection. However, none of these books contains a critical review of the effectiveness of the technologies. One instrument included in the publication is a portable, handheld, ion mobility spectrometry chemical agent monitor with moderate to high selectivity, but only when used in open spaces, far from vapor sources such as smoke, cleaning compounds, and fumes. This would seem to make it useless in the battlefield. Another listed chemical agent monitor has a below 5% false positive rate. With one in 20 false positives, no one could reasonably act upon an alarm. 

A caveat to the above discussion is that some sensor system implementations utilize low cost, low selectivity sensors deployed at a high rate across the area of interest. Higher cost, high-selectivity instruments are deployed to less locations across the area of interest. In the event of an alarm by one of the higher selectivity instruments, the lower selectivity instruments can be used to pinpoint the location of the vapor cloud and perhaps its origin. Thus, it may be that not every chemical sensor needs perfect selectivity, but clearly a high rate of false alarms can rarely be tolerated. 

The operation of many safety related systems is dependent upon a number of interacting parameters. Frequently these parameters must be tuned to the particular operating environment to provide the best possible performance. We focus on the Short Term Conflict Alert (STCA) system, which warns of airspace infractions between aircraft, as an example of a safety related system that must raise an alert to dangerous situations, but should not raise false alarms. Current practice is to tune by hand the many parameters governing the system in order to optimise the operating point in terms of the true positive and false positive rates, which are frequently associated with highly imbalanced costs. 

This method separates the window cells into a leading and a lagging part. Before the mean values of these parts are averaged, they are weighted by the factors a and (3. Optimum values for a and /3 are calculated in accordance with the level of interference of present targets, so that a constant false alarm rate and a high detection probability can be guaranteed. 

Generally, the performance characteristics of greatest interest for an explosives detection system are sensitivity, selectivity, and response time. As used here, sensitivity is the ability to detect the target analyte in extremely small concentrations, while selectivity is the ability to distinguish the target analyte from other materials that may be present. In combination, good sensitivity and selectivity mean a high probability of detection when the analyte is present and a low false alarm rate when the analyte is not present. 

There have been many quite useful discoveries in chemistry and chemical engineering over the years that have been used for detection applications. The first example is plasma chromatography, otherwise known as ion mobility spectrometry. In the 1980s this technique became the method of choice for detecting chemical warfare agents and was used by soldiers in Desert Storm with unfortunate results. Official reports tell that the rate of false alarms for these instruments was so high that soldiers became desensitized to real hazards. One infantry battalion eventually turned their alarms off. Much of the Gulf War Syndrome may well have been caused because ion mobihty spectrometry was oversold as a detection technique. 

Figure C.l outlines how a simulation might be constructed to explore this scenario. The goal is to calculate delays that would be introduced into a deployment. The simulation begins with the question of whether any indications and warnings (I W) of an attack are available before the attack actually begins. Questions related to where such I W might originate are discussed below. Sensors will produce a finite number of false alarms, which require some time, tFA, to resolve. One can anticipate that there will invariably be some time delay, associated with the total number of false alarms and the time needed to resolve them. For sensors with high false-alarm rates, t. goes up.

A subset of the ARL results from this study listed in Table 2.3 indicate that the in-control ARL are very sensitive to the presence of autocorrelation, but the detection capabilities of CUSUM and EWMA for true shifts are not significantly affected. In the absence of autocorrelation, the ARL(O) for CUSUM is 465 and that for EWMA is 452. The ARL(O) for low levels of autocorrelation 4> = 0.25) are 383 and 355, respectively, and they drop drastically to 188 and 186 for high levels of autocorrelation (f> = 0.75), increasing the false alarm rates by a factor of 2.5. 

---