Vdt - Lighting Research Center · EXPERIMENT - Lamp Life Test As part of our ongoing lamp-ballast system compatibility research, several F32T8 fluorescent lamp and ballast systems

1

IES Paper # 48

A LAMP LIFE PREDICTOR FOR FREQUENTLY SWITCHED INSTANT-START FLUORESCENT SYSTEMS

N. Narendran, T. Yin, C. O’Rourke, A. Bierman and N. Maliyagoda

Lighting Research Center Rensselaer Polytechnic Institute, Troy, NY 12180.

Tel: (518) 276 4803; Fax (518) 276 4835 ABSTRACT Although T8 fluorescent lamp life is typically rated at 20,000 hours, it can significantly vary depending on the ballast product. The current industry practice for testing fluorescent lamp life requires these lamps to be subjected to 3-hours on/20-minutes off cycle, which is also known as the standard cycle for life testing. This method requires over 2 years of testing, which is expensive and also delays the development of new products. Therefore, researchers have been trying to expedite the procedure for lamp life testing by increasing the switching frequency. However, none of these attempts have shown much promise for correlating the lamp life on faster cycles to the lamp life on the standard cycle. Alternatively, researchers have been trying to correlate certain starting and operating electrical parameters of the lamp-ballast system to lamp life. As an example, it has been shown that the ratio of the hot cathode resistance to cold cathode resistance correlates well with lamp life for rapid-start ballast fluorescent systems. Instant-start ballasts comprise a larger market share than rapid-start ballast. A literature survey shows that at the present time there is no such parameter available for predicting lamp life of instant-start systems. It is also shown in literature that within a range, lamp life shortens with increasing switching frequency. At very fast switching frequencies such as the 5-minutes on/5-minutes off cycle, the major portion of the electrode damage takes place due to starting, which ultimately leads to lamp failure. At standard operating cycles, the electrode damage takes place during starting and operating periods. To predict lamp life on any switching cycle, one has to know the amount of electrode damage that takes place due to starting and the amount of electrode damage that take place due to normal operation. This study deals only with starting effects. The goal of this initial study is to analyze the starting electrical properties and identify a parameter that correlates well to high-frequency switching life, which provides insight to the amount of electrode damage that takes place during starting. It is shown here that the time integrated value of the lamp voltage over the starting

period, Vdtstart

�0

, correlates well with high-frequency switching life for the lamps tested. The details of the

experimental study and the results are explained in this manuscript. INTRODUCTION Of all the different ballast types available for linear fluorescent lamps, instant-start ballasts are the most commonly used, over 70%, especially with F32T8 lamps [1]. The widespread use of instant-start ballasts is mainly attributed to lower cost, lower energy consumption, and its ability to start lamps without delay. Over the past several years, the number of manufacturers supplying instant-start ballast products to the marketplace has been growing steadily. Although having many choices for ballast products is a positive trend for the fluorescent lamp industry, it is confusing for the end users who have to select a suitable lamp and ballast system for their target applications. Past research has shown that fluorescent systems with incompatible lamps and ballasts result in shortened lamp life [2-8]. Therefore, it is very important to appropriately combine a lamp with a ballast to obtain longer lamp life. Traditionally, lamp life is determined by subjecting fluorescent lamps to a 3-hours on/20-minutes off cycle, which is known as the standard cycle for life testing. Testing fluorescent lamps that have median life of 20,000 hours, on the standard cycle can take up to 2.5 years. This is expensive and also delays the development and introduction of new products. Therefore, researchers have been trying to expedite the procedure for lamp life testing by increasing the switching frequency [9]. However, none of these attempts have shown much promise for correlating the lamp life on faster cycles to the lamp life on the standard cycle. Alternatively, researchers have been trying other methods such as measurement of certain electrical parameters, use of laser induced fluorescent (LIF) or use of x-ray fluorescence measurement techniques to study coating loss from the

2

electrodes, which is then used to predict lamp life [3-4, 6-8, 10-17]. Some of these approaches have shown limited success while others have shown greater success in predicting lamp life. As an example, it has been shown that the ratio of the hot cathode resistance to cold cathode resistance, Rh/Rc, correlates well with lamp life for rapid-start ballast fluorescent systems [4,8,18]. In 1998, Ji et al systematically studied the Rh/Rc for different lamp/ballast combinations in the market place and showed that Rh/Rc correlated well with lamp life [8]. D. Hitchcock in 1983, conducted a study with instant-start fluorescent systems to relate operating electrical parameters to lamp life [7]. However, at present there isn’t any published information that relates starting electrical parameters to lamp life for instant-start systems. Therefore, an experimental study was conducted to investigate the possibility of identifying a lamp life predictor for fluorescent lamps operated on instant-start ballasts by analyzing the starting electrical parameters. In practice, fluorescent lamps undergo different on-off cycles. As an example, lamps used in open-plan office spaces, on an average, can be on for 18 hours and off for 6 hours in a day [19], whereas in conference rooms, on an average they are on for 5 hours and off for 19 hours [20]. End of life for a fluorescent lamp is defined when the lamp fails to start. Usually, a fluorescent lamp reaches end of life when the emissive coatings of its electrodes evaporate or sputter away completely. There are times, especially with instant-start systems, when a lamp would operate even after the emissive coatings have disappeared. The authors of this paper assume that the time of operation after complete loss of emissive coatings is small. The depletion of emissive coatings takes place during starting and operating periods. The rate at which the emissive materials are lost during the starting process is different to the rate at which they are lost during operation. Assuming that the lamp life is determined by the coating losses taking place during starting and operating periods only, the total coating loss, L in percentage, can be written as,

L = Ns . DD + Nhrs . EE (1) where, Ns and Nhrs represent the number of starts and number of operating hours respectively, D and E represent the coating loss per start and coating loss per hour of operation respectively. When L is equal to 100% the lamp reaches end of life. The authors suspect that there may be other factors that contribute to lamp life, but those factors may be small compared to the ones shown in equation (1). The life of a fluorescent lamp that operates continuously for a long period depends mainly on the damage that occurs during lamp operation and the life of frequently switched fluorescent lamps depends mainly on the damage that occurs during lamp starting. Assuming that D and E are the two dominant factors and knowing their values for a particular lamp-ballast system one can estimate the life of the system for any on-off cycle. Subjecting similar lamp and ballast systems to two different on-off life testing cycles, as an example 5-minutes on/ 5-minutes off cycle and 3-hours on/ 20-minutes off cycle, one can determine D and E� $ sample estimate of D and E values for six different instant start ballast and lamp combinations is illustrated in Table 2. These values were calculated from the results of a life study experiment that took 3 years to complete. This process takes a very long time. Ideally, if one can determine the percent damage due to starting, D� and operating, E� by measuring certain electrical parameters during these two periods, it would significantly reduce the time and cost involved in life testing fluorescent lamps. The end use community can significantly benefit by using this method to select a lamp-ballast system optimum for their target application. Likewise the manufacturing community can significantly benefit by using this method to rapidly improve product performance. EXPERIMENT - Lamp Life Test As part of our ongoing lamp-ballast system compatibility research, several F32T8 fluorescent lamp and ballast systems were put on a life test rack in 1996. In that study, several lamp and ballast combinations were subjected to two different life-testing cycles, namely 3 hours on/20 minutes off and 5 minutes on/5 minutes off. That previous study was aimed at collecting life data for different starting methods with concentration on rapid-start systems. However, two groups of instant-start systems were included in that study for comparison with rapid start systems. The two instant-start electronic ballasts, named B3 and B9, had significantly different life on the two switching cycles and therefore they are the ones of interest for our study presented in this manuscript. Table 1 summarizes the median life data for the different lamp and ballast combinations subjected to the 5 minutes on/5 minutes off and 3 hours on/20 minutes off cycle tests.

3

In addition, Table 1 summarizes the number of lamp-ballast systems used in each case. Table 2 illustrates the calculated D and E values of the six different systems. Prior to starting the life test in 1996, LRC researchers measured certain ballast parameters for the instant-start lamp and ballast systems. Some of the measured parameters were lamp power, open circuit voltage, current-crest factor, and ballast factor [8]. Most of these parameters show very little difference and do not explain the difference in life data for the lamps operating on the two different ballast products. In one case, the lamp life on B9 ballast exceeded the lamp life on B3 ballast by 290% when they were subjected to the 5-minutes on/5-minutes off cycle. Since the above measured electrical parameters were not capable of explaining the lamp life differences of the instant-start systems, further research was needed to identify electrical parameters that correlate well to lamp life. Understanding the reason for the difference in lifetime and identifying a parameter that correlates well to lamp life for high-frequency switching cycles became the goal for our study presented in this manuscript. This study is limited to high-frequency switching cycles because complete life data was available only for the fast cycle when this study was started.

Ballast Type

B3 B9

Lamp A B C A B C Median Lamp Life (hrs) for 5-min on/5-min off cycle.

1,487

n=12

711

n=12

981

n=12

1,597

n=12

1,191

n=12

2,064

n=12

Median Lamp Life (hrs) for 3-hrs on/20-min off cycle.

24,737

n=4 *

13,842

n=6

14,162

n=6

21,120

n=6

19,536

n=6

>25,000

n=6

Table 1: The median lamp life data for instant start systems, n refers to the number lamps used in each case. * Due to early ballast failure for two pairs, the sample size is reduced to 4 from 6.

Ballast Type

B3 B9

Lamp A B C A B C

DD (%/ start)

5.42 x10-5

11.44 x10-5

8.13 x10-5

4.96 x10-5

6.76 x10-5

3.81x10-5

EE (%/hr)

2.24 x10-5

3.41 x10-5

4.35 x10-5

3.08 x10-5

2.87 x10-5

2.70 x10-5

Table 2: Calculated D and E values. EXPERIMENT – Starting Electrical Parameter Measurement Figure 1 illustrates the schematic of the experimental setup utilized for measuring the starting electrical parameters. All ballasts used in this study are two lamp ballasts. The test setup and procedures conformed to American National Standard Institute (ANSI) requirements [21,22]. A digital thermometer was placed at a point not more than 3 feet away from the lamps and at the same height as the lamps to monitor the ambient temperature. The ambient temperature throughout the measurement period was in the range of 25qCr1qC (77qFr1.8qF). The starting aid distance, 0.5 inch, was kept constant for all the measurements.

4

An ac power source supplied the necessary input voltage, 120 volts r 0.12 volts, at 60 Hz frequency to the experimental setup. A four-channel digital storage oscilloscope was utilized for capturing starting voltage and current waveforms from the blue and red wire lamps. Two differential voltage probes (attenuation ratio – 1/200) were used for voltage measurements and two current monitors (0.1 Volts/Amp) were used for current measurements. When power was applied to the ballast the fluorescent lamps were started and the digital storage oscilloscope captured all the starting waveforms. The digital storage oscilloscope was set to trigger at 160V dc and the sampling rate was set at10 million samples per second. The acquired waveforms were directly transferred to a computer via IEEE-488 interface and a commercial software package, LabVIEW, was utilized to further analyze the data. A total of 6 lamp-ballast systems were studied in this experiment. The ballasts are 2-lamp ballasts and each system had 3 sets of lamp samples and 3 ballast samples. All ballasts and lamps used in this experiment were from the same manufacturing batch as the ballasts and lamps that were life tested. The lamps were seasoned for 100 hours prior to making measurements. The same sets of lamps were used with the various ballasts. Two lamps were connected to the same ballast for which 4 repeated measurements were made on each of the 6 ballast samples. Altogether 144 measurements were made for all the lamp-ballast combinations.

DATA ANALYSIS Figure 2 illustrates the ANSI described starting current waveform for the instant-start fluorescent system [22]. As shown in the figure 2, t0 is the time when power is applied, t1 is the time when glow current first appeared and t3 is the time when peak of the first half cycle is at least 90% of the waveform and is sustained at that value. From ANSI definitions the starting time, t, can be written as

13 ttt �� . (2)

Figure 1: Experimental setup utilized for measuring the starting electricalparameters

Oscilloscope

Regulated AC Power Supply

Line

/

1

Ballast

Lamp 2

Lamp 1

&+ &+ &+ &+

Differential Probes

5

However, in our experiment it was noticed that the starting waveform looks very much different than what ANSI illustrates. Figure 3 illustrates a sample current waveform obtained from our experiment. The ANSI illustration of the waveform may fit typical 60Hz magnetic ballast operation, but it does not reflect the more complex operation on a high frequency ballast.

Initially, this discrepancy posed a challenge in deciding how to measure the starting time for the lamp and

ballast systems. Figure 4 illustrates a sample starting voltage waveform and the corresponding current

waveform for lamp-ballast system using B9 ballast. Likewise, Figure 5 illustrates a sample starting voltage

waveform and the corresponding current waveform for lamp-ballast system using B3 ballast. From figures

4 and 5 it can be seen that lamps on ballast B3 takes longer to stabilize compared to the lamps on ballast

B9. It was noted in many cases that the starting current and the starting voltage didn’t stabilize at the same

time. In our analysis the starting time was taken to be the time needed for both, lamp current and lamp

voltage, to stabilize. Typically, the current waveform took longer than the voltage waveform to stabilize.

Starting time was calculated according to equation 2, where t1 is the time when the first glow current

Figure 2: ANSI recommend starting current waveform for theinstant-start fluorescent system [22]

Figure 3: A sample current waveformobtained from our experiment.

6

appeared and t3 is the time when the lamp current stabilized. The current was considered stabilized when

the rms value after time t3 did not vary more than 10 percent from the rms value of the operating current. It

also became apparent that the starting time varied every time a system was started. Therefore, for every

lamp-ballast system, 8 repeated measurements were made, with 1 hour off period in between

measurements. The average value of these 8 measurements was taken as the starting time along with its

standard deviation value for a given lamp ballast combination.

Figure 5: Sample starting voltage waveform and the corresponding currentwaveform for a lamp-ballast system using B3 ballast

Figure 4: Sample starting voltage waveform and the corresponding currentwaveform for a lamp-ballast system using B9 ballast

7

The other starting parameters such as the root mean square (rms) values of the lamp voltage and current applied over the starting time period were calculated once the starting time was decided. The starting power, P, is the time-averaged value of the instantaneous power, calculated from the instantaneous values of V and I, over the starting time period. Correlation between the starting parameters and system life was explored. Figures 6a and 6b illustrate the correlation between lamp life when frequently switched and starting time and starting voltage. In both these plots the y-axis represents the lamp life and the x-axis represents the corresponding starting parameters. In addition best fit regression lines are drawn through the data points and the corresponding variance values, R2, are utilized to estimate how well the parameters correlate to lamp life. It can be seen that the starting time and starting voltage correlate well with lamp life with respective correlation values 0.89 and 0.91 [2]. However the starting current and starting power showed much lower correlation. From this analysis it appears that starting time and starting voltage are two parameters that can be used for predicting lamp life. Of the two parameters, starting time is not a physical cause for electrode damage, but starting voltage could influence physical damage. Furthermore, voltage is not constant throughout the starting time, and the electrode damage depends on the magnitude of the starting voltage. Therefore, it appears that integrating

the starting voltage, Vdtstart

�0

, over the starting time would better represent the cumulative damage caused

by the starting voltage during the starting period.

The Vdtstart

�0

value which is the area under the curve of the voltage waveform, was calculated for each

system. The data acquisition software seperated the cathode and anode half cycle voltage waveforms and

took the absolute value of the voltage when calculating Vdtstart

�0

. Since lamp failure would take place when

either of the two electrodes lost all of its emissive material, the larger Vdtstart

�0

value of the two electrodes

was used for analysis. For every ballast the Vdtstart

�0

values from each lamp were averaged.

Figure 7 illustrates the plot of lamp life versus Vdtstart

�0

. It should be noted that V inside the integral is not

the rms starting voltage, instead it is the instaneous value of the starting voltage at each time value. A best-

fit regression line shows a very high correlation between Vdtstart

�0

and lamp life, with a R2 value of 0.94

[2].

8

(a)R

2 = 0.89

0

500

1000

1500

2000

2500

0 10 20 30 40 50 60

Starting time (ms)

Lif

e (h

rs)

(b) R2 = 0.91

0

500

1000

1500

2000

2500

0 50 100 150 200 250 300 350 400

Starting Voltage (V)

Lif

e (h

rs)

Figure 6: Plot of lamp life versus a) starting time; b) starting voltage [2]

Figure 7: Plot of lamp life versus the time integrated value of the lamp

voltage over the starting period, Vdtstart

�0

.[2].

R2 = 0.94

0500

1000150020002500

0 2 4 6 8 10

Vdt (Volts.sec)

Life

(hrs

)

Vdtstart

�0

(V.S)

Starting Time (ms)

R2 = 0.91

9

DISCUSSIONS The electrode structure of a fluorescent lamp has an emissive coating. As mentioned earlier, once the emissive coating is totally lost the lamp is close to end of life. The lamp electrode performs the function of transferring charges between the external circuit and the positive column. In an a-c discharge, the electrode structure performs three functions, supplying electrons to the discharge and collecting ions on the cathode half-cycle, and receiving electrons on the anode half-cycle. A cathode coating that reaches thermionic emission rapidly will supply electrons to the discharge promptly and minimize cathode erosion during starting. This is important since the high ion bombardment of the cathode during starting, when it is not at emission temperature, will greatly erode the cathode coating [23]. The ANSI recommends a starting time of less than 100ms for instant-start systems. If all other parameters are equal, shorter starting time will result in less sputtering damage to the electrode emissive material. Therefore, the lamp life will be longer. This explains the good correlation between lamp life and starting time, t, observed in our experiment. The results show that longer starting time results in shorter lamp life. When the electrode is negative with respect to the surrounding discharge, emitted electrons are accelerated through the cathode sheath, into the negative glow of the lamp discharge due to the potential difference across the lamp. These electrons collide with atoms of mercury and argon in the negative glow region, giving up their energy to produce mercury ions. The ions that reach the cathode strike it with energy equal to the kinetic energy gained through the cathode fall region. Therefore, the ions closer to the cathode will typically have less kinetic energy and the ions further away will have greater kinetic energy. During the starting period of the instant-start system the major portion of the starting voltage is cathode-fall voltage and typically this is higher than the cathode-fall voltage during lamp operating period. When the cathode fall voltage is high, the incoming ions are able to physically eject material from the cathode surface, resulting in severe sputtering, because they possess higher kinetic energy that results in higher momentum [3,24]. Sputtering is the main reason for the loss of emissive material during starting. This explains the good correlation between the lamp life and lamp starting voltage, V, observed in our experiment. The results show that higher starting voltage results in shorter lamp life. Although starting time and starting voltage showed good correlation, 0.89 and 0.91, with lamp life,

respectively, the time integrated starting voltage over the starting period, Vdtstart

�0

, showed the best

correlation, 0.94, with lamp life [2]. In addition, as seen in figures 6a & b some data points overlap because

of the variations in starting time and starting voltage. However, in the case of lamp life versus Vdtstart

�0

,

the overlap between Vdtstart

�0

values is minimal. As mentioned earlier, during the starting period significant

portion of the starting voltage is cathode fall voltage. Integrating the instantaneous voltage values over the starting period provides a better estimate of the accumulated electrode damage since it accounts for the voltage fluctuations that are directly related to the energy gained by the ions that cause the sputtering

damage. Therefore, Vdtstart

�0

is a good predictor of lamp life for frequently switched instant start

fluorescent systems. The correlation between lamp life and time integrated value of the starting current over the starting period,

Idtstart

�0

, is low. At first glance it looks surprising that starting current doesn’t correlate well with lamp life,

since high current means more electrons in the discharge region of the lamp, which means more ions and

10

finally more damage to the electrode. However, there are many factors that influence the number of ions that bombard the electrode. These factors include the probability that an emitted electron produces an ion in the negative glow region, the probability that the ion will reach the cathode and the energy distribution of the ions [25]. Therefore, a high starting current does not necessarily mean high ion bombardment. Also, during starting the current is carried by the ions and ions near the cothode will count as current but may not cause much damage to the cathode for the reasons explained earlier. This explains the poor correlation between the starting current and lamp life. Since starting current has poor correlation, when multiplied with starting voltage to produce power P, it too gives a poor correlation with lamp life. SUMMARY As part of an ongoing lamp-ballast system compatibility research, in 1996 several 4-foot F32T8 fluorescent lamp and ballast systems were subjected to two different life-testing cycles, namely 3 hours on/20 minutes off and 5 minutes on/5 minutes off cycles. The goal of the 1996 study was to gain better understanding of lamp life for different starting methods. Although, the study mainly dealt with rapid-start systems, two groups of instant-start systems were included in that study for comparison. The two instant-start electronic ballasts showed significantly different lifetimes than each other on both switching cycles. Understanding the reason for the differences in lifetime and identifying a parameter that correlates well to lamp life for fast switching cycles became the goal of our study presented in this manuscript. This study was limited to high-frequency switching cycles, where most of the electrode damage takes place due to starting. The main reason for selecting only the rapid cycle test data is, when this study was started complete life data was available only for the fast cycle. It is shown in this study that the time integrated value of the lamp voltage

over the starting period, Vdtstart

�0

, correlates well with fast switching life for the lamps tested.

Having only two ballast types is a limitation of this study. Further, investigation is needed with many more ballast types to confirm the results observed in this study. A second phase of this research is currently

underway and one of the goals is to use many different instant-start ballasts and verify that Vdtstart

�0

correlates well with high-frequency switching life. As mentioned earlier, high-frequrncy switching lamp life cannot be correlated to standard cycle lamp life. To be able to predict lamp life on any cycle one has to know the amount of electrode damage due to starting and operating seperately. Therefore, another goal for the second phase study is to analyze the operating parameters and identify a parameter that can predict emissive coating loss due to operating. The results of the second phase study will be published at a later time when it is completed. ACKNOWLEDGEMNTS The authors gratefully acknowledge the support of ESEERCO and NYSERDA. The authors are grateful to Dr. Victor Roberts for his valuable discussions. REFERENCES 1. U.S. Bureau of the Census, Fluorescent lamp ballasts, MQ335C, MQ36c(99)-1, May 1999. 2. T. Yin, “Predicting life for instant start ballast/fluorescent lamp systems on short cycles”, M.S. Thesis,

1999. 3. Covington E. J., Life Prediction of Fluorescent Lamps, Journal of IES, March 1971 4. Hammer E. E., Photocell Enhanced Techniques For Measuring Starting Electrodes Temperatures of

Fluorescent Lamps, IEEE annual meeting, October 1997 5. Hammer E. E., Fluorescent System Interaction with Electronic Ballasts, Journal of IES, Winter 1991 6. Hammer E. E., Comparative Starting-Operating Characteristics of Typical F40 Systems, Journal of

IES, Winter 1989 7. Hitchcock D. E., High Frequency Characteristics of 32W T8 Lamps, Journal of IES, October 1983

11

8. Ji,Y., Fluorescent Lamp/Ballast Systems Research Report, 94-40, October 1998 9. Vorlander, F.J. and Raddin, E.H., 1950, “The effect of operating cycles on fluorescent lamp

performance”, Illuminating Engineering, January 1950. 10. Misono,K. Cathode Fall Voltage of Low-current Fluorescent Lamps, Journal of IES, P.108-115,

Summer 1991 11. Garbowicz, G. Hybrid Ballast II- High Efficient Type Ballasts for 32W-T8 and 34W-T12 Lamp

Systems, Journal of the Illuminating Engineering Society, Vol.23, No.1, P.22-30, 1994 12. Bhattacharya, A.K., Measurement of Barium Ion Density in the Vicinity of Fluorescent Lamp

Electrodes, Journal of Applied Physics., 65(12),15:4602-4607, 1989 13. Moskowitz,P., Fluorescent Lamp Electrode Diagnostics by Laser Induced Fluoresce, Paper presented

at the 6th international symposium on the Science & Technology of Light Sources, 1992 14. Hammer, E.E. Cathode Fall Voltage Relationship with Fluorescent Lamps, Journal of the Illuminating

Engineering Society, Vol.24, No.1, P.116-122, 1995 15. Verderber,R.R., et al. Life of Fluorescent Lamps Operated at High Frequencies with Solid-state

Ballasts, Paper presented at the IEEE Industrial Application Society annual meeting, Toronto, Canada, 1985

16. Mortimer, G.W, Real-time measurement of dynamic filament resistance, Journal of IES, P.22-28, Winter 1998

17. Watanabe, Y. and Murayama, S., Cathode Fall Characteristics of Fluorescent Lamps under High Frequency Operation, Journal of Applied Physics 32:3593-3600, 1993

18. Stephens D. and Wolfman H., Procedure to Measure Filament Heating in a Fluorescent Lamp during Rapid Start Heating, Handbook of Chemistry and Physics, 73rd edition, Chemical Rubber Company, 1993

19. DELTA Portfolio, Vol.1, Issue 5, 1996, Lighting Research Center, Rensselaer Polytechnic Institute, Troy, NY

20. D. Maniccia, A. Tweed, B. Von Neida, and A. Bierman, “The effects of changing occupancy sensor timeout setting on energy savings, lamp cycling, and maintenance costs”, Illuminating Engineering Society of North America Annual Conference, 2000.

21. American National Standard Institute, American National Standard For Lamp Ballasts—Lamp Frequency Fluorescent Lamp Ballast, ANSI C82.1, 1997

22. American National Standard Institute, High-frequency fluorescent lamp ballasts, ANSI C82.11, 1993 23. Toomey C. L., Effects of Electrode Structure on Lamp Performance, Journal of IES, June 1963 24. Soules T.F., Ingold J. H., Bhattacharya A. K., and Springer R. H., Thermal Model of The Fluorescent

Lamp Electrode, Journal of IES, Summer 1989 25. Waymouth, J. F., Electric Discharge Lamps, Cambridge, MA, MIT Press, 1971