1316 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 53, NO. 4, AUGUST 2004 On-Line Fuel Identification Using Digital Signal Processing and Fuzzy Inference Techniques Lijun Xu, Member, IEEE, Yong Yan, Steve Cornwell, and Gerry Riley Abstract—This paper presents a novel approach for on-line fuel identification using digital signal processing (DSP) and fuzzy inference techniques. A flame detector containing three photodi- odes is used to derive multiple signals covering a wide spectrum of the flame from infrared to ultraviolet through visible band. Advanced digital signal processing and fuzzy inference techniques are deployed to identify the dynamic “fingerprints” of the flame both in time and frequency domains and ultimately the type of coal being burnt. A series of experiments was carried out using a 0.5- combustion test facility operated by RWE Innogy plc, U.K. The results obtained demonstrate that this approach can be used to identify the type of coal being burnt under steady combustion conditions. Index Terms—Combustion, digital signal processing (DSP), flame detector, fuel identification, fuzzy logic, soft-computing. I. INTRODUCTION P OWER plants are increasingly burning a more diverse re- source of coals under tighter economic and environmental constraints. Experience has shown that boiler optimization packages can help plant operators to optimize the combustion process and hence improve its efficiency for a given type of fuel [1]. Power stations that burn consistent coal diets have demonstrated that the coal combustion process can be opti- mized in terms of reduced emissions and carbon-in-ash levels. However, a power station can have a wide range of coals in its stock but what type of coal is being fired at any moment is often unknown and even unpredictable. The application of the optimization packages is thus seriously limited by the wide variation of the coal diet. Therefore, on-line fuel identification at a power station where a wide range of coals is used would improve the performance of the optimization packages leading to increased combustion efficiency and reduced pollutant emissions. On-line coal analyzers operating on radiometric, microwave and infrared methods are available on the market [2], [3]. Pas- sive tagging techniques have also been adopted for on-line coal tracking [4]. However, these systems are very expensive and require complex installation as their operations involve either taking samples from the coal feeding system or detecting tracer particles that have been added into the coals. Manuscript received June 15, 2003; revised March 24, 2004. This work was supported by the U.K. Department of Trade and Industry by a Grant-in-aid. L. Xu and Y. Yan are with the Instrumentation and Embedded Systems Re- search Group, Department of Electronics, University of Kent, Canterbury, Kent CT2 7NT, U.K. (e-mail: l.xu@kent.ac.uk; Y.Yan@kent.ac.uk). S. Cornwell and G. Riley are with the RWE Innogy plc, Swindon SN5 6PB, U.K. Digital Object Identifier 10.1109/TIM.2004.830573 Fig. 1. Block diagram of the measurement system. Flame radiation covers a wide spectrum from infrared (IR) to ultraviolet (UV) through the visible band [5], [6]. Different coals may be different in physical ingredients and chemical el- ements. Flames generated by different coals under steady com- bustion conditions (coal feeding, oxygen feeding, etc.) possess different features both in time and frequency domains which can be extracted as unique signatures to identify the types of coal. In this paper, a novel approach is reported that modifies the flame detectors existing on all power stations and utilizes advanced signal processing and fuzzy inference techniques to identify the coal or “family” of similar coals that is being burnt. This new approach, once fully developed, has obvious advantages over other systems including cost-effectiveness, easy installation, and low maintenance requirement. II. METHODOLOGY In a power plant, the routine operation of a boiler generally requires steady combustion conditions such as a certain quan- tity of coal feeding, coal-air ratio, etc. Under these conditions, different coals are expected to produce flames with different dis- tributions of radiation energy across the electromagnetic spec- trum. A specially designed flame detector with the same instal- lation specifications as the traditional one is used to derive mul- tiple signals covering different bands of the flame spectrum [7]. Digital signal processing (DSP) and fuzzy inference techniques are deployed to identify the dynamic “fingerprints” of the flame both in time and frequency domains and ultimately the type of coal being burnt. A. System Description The monitoring system consists of a flame detector, a signal conditioning circuit, an I/O interface and a PC-based signal pro- cessing unit. Fig. 1 shows the structure and main constituent el- ements of the system. The flame detector is designed in order to 0018-9456/04$20.00 © 2004 IEEE