Introduction
Fire and smoke kill more people every year than many other forces. While controlled fire serves us in so many instances, uncontrolled fire can be of harm, however, the rapid detection of fire and its control can save lives and property damage worth millions. Conventional and addressable are two main types of fire alarm systems, but unfortunately, these fire alarm systems often generate false alarms. The ratio of false alarm is higher in conventional alarm systems compared to addressable, but addressable alarm fire systems are more expensive. The most likely cause of a false warning is different for distinct types of detection systems, such as a smoke sensor often being activated falsely due to an environmental effect. So, there is a need for a cost-effective multi-sensors expert alarm system that is artificially trained and assists FDWS (fire detection and warning system) to make the right decisions and to reduce the number of false alarms. False alarm warnings are so common that London fire brigade alone is called out almost every 10 min to attend a false alarm causing them a loss of about £37 million per year. To achieve the aforementioned goal, in this paper, we introduced a home-based FDMS that uses a microcontroller Arduino UNO R3 (Arduino, Somerville, TX, USA) based on the atmega328p. It is easily available and programmed using the Arduino Software (IDE) with a set of cost-effective sensors. The proposed solution effectively uses a smoke sensor with flame sensors with a particular increase in room temperature; to further investigate the true presence of fire and to avoid false alarm, the FDWS is trained with a neuro-fuzzy designer. The purpose of this intelligent fire alarm system is to sense true occurrences of fire, alert the proper authorities, and notify the occupants via GSM to take necessary action immediately.
A false alarm can burden the fire brigade and can turn out to be a costly event; so many studies conducted to reduce them. Previous studies proposed different methods such as autonomous firefighting robots, fire alarm systems with notification appliances, and wireless warning systems. Fire alarm systems with notification appliances can be costly because they use visible and audible stimuli to notify residents. The primary objective of this paper is to develop a reproducible and economical solution with minimum false alarms and a system that alerts via GSM (global system for mobile communication). The innovative idea is to use neuro-fuzzy logic to design a smart alarm system. Our proposed system is ANFIS-simulated in MATLAB environment; the obtained results show effectiveness and the robustness with good performances compared with the FIS method (in Section 3). The ANFIS idea was originally proposed by Jang in 1993. Typically, an ANFIS is a combination of a neural network and a fuzzy inference system (FIS) and is effective in making decisions.
Related Works
This section discusses different AI techniques and other fire detection methods used in the past to mitigate risks of fire by early detection and reduce false warnings, but our main focus is ANFIS technology. Efforts were made for early fire detection and risk mitigation. Diverse technologies developed by researchers have been used such as fuzzy logic, neural networks, video-based techniques, Image Processing color-based fire detection methods, etc. Early Fire detection always has been an important research Topic for researchers. The idea of using multiple sensors was proposed by Faisal et al. The proposed wireless sensor network (WSN) consists of different sensors that share a single wireless network and used GSM. The proposed system results were tested in a smart home to reduce false warnings. Elias et al. also provided a solution using wireless sensor network that was embedded in a micro-controller board for fire hazard detection and fire monitoring purpose.
Hamdy et al. Built a “Smart Forest Fire Early Detection Sensory System (SFFEDSS)”, by combining the wireless sensor networks and artificial neural networks for the detection of forest fire.
Yu et al. collected the sensor readings for smoke intensity, humidity, temperature to use it in fire detection using Feed-forward neural network approach. The disadvantage of a Feed-forward approach is it demands high processing at the node level resulting in a large amount of power consumption which reduces the lifespan of the node. Also, cluster head destruction in the fire badly affects the robustness of the system.
A system presented by Vikshant et al. works for detection of forest fire by combining wireless sensor networks (WSNS) with fuzzy logic. Multi-sensors technology is used for detecting fire chances and early fire detection. Information gathered from different sensors such as heat, humidity and CO density light, will be sent on the cluster head using event detection mechanisms. Multiple sensors used to detect fire probability and direction are embedded in each node to reduce the false alarm rate and improve the efficiency. A simple way to detect fire developed by Muralidharan et al. using Multiple sensors with the implementation of fuzzy logic and presented the obtained results in MATLAB.
In 2017, Yu-Liang Hsu. developed a multi-sensor data fusion technology with artificial intelligence, wearable intelligent technology, and sensor fusion technology that can control home appliances and locate the position of home residents. It works in indoor environments. Similarly, a system was developed by Mirjana et al. which used an IoT concept for determining true fire presence according to the situation.
Robert et al. introduced a system using Arduino microcontroller and fuzzy logic technology in search of fire detection in automobile and to reduce its damage due to fire. Different sensors like temperature sensors, smoke sensors, and flame sensors were used. This system was tested on an average-sized car with 2 kg cylinder mounted behind the passenger’s rear seats.
J Olivares-Mercado proposed a method of early fire detection by analyzing visual smoke characteristics such as color, dynamic texture, gray tones, etc. The system was tested using standard videos containing smoke.
JH Park. proposed an early fire detection system for smart cities with a multifunctional artificial intelligence framework. The artificial intelligent framework includes an adaptive fuzzy algorithm, machine learning algorithms and Direct-MQTT based on SDN.
In this paper, ANFIS technology is used to design a fire detection control system and reduce false alarms. ANFIS technology has been used in mobile robot navigation, healthcare monitoring systems, air conditioning control, flood susceptibility modeling, and many other applications. In recent times, fiberoptic sensors were used for structural fire engineering, however, there is a need for true fire identification. IoT is successfully being used to achieve accuracy and efficiency in modern smart systems and has provided positive results as well. This success of IoT was the motivation to design the proposed a smart and intelligent system for fire monitoring. In the rest of the paper, Section 2 discusses the used approach and architecture of the proposed ANFIS-based decision support system for early fire identification and details of implementation. Section 4 discusses experimental settings with their results and this work is concluded in Section 5.
- Materials and Methods
2.1. Adaptive Neuro-Fuzzy Inference System (ANFIS) Architecture
For generating an intelligent fire detection system that can monitor the parameters required for the actual presence of fire so that a false alarm can be decreased up to a minimum level, a combination of two important technologies fuzzy logic and artificial neural network (Ann) called the adaptive neural fuzzy interface system (ANFIS) can logically generate fuzzy rules according to training data to make the system robust. A fire detection system is developed using this aforementioned technology and presented in this paper to find the probability of fire. The ANFIS neural network works until the output matches the desired value for the given input. So, considering these abilities, an adaptive neuro-fuzzy interference system is used for detection of fire.
ANFIS is a five-layer architecture that was developed in the early 1990s. The first layer is called the input layer. The second layer of ANFIS, called inputmf, is a fixed input membership function layer. The third layer depicts norms. The fourth layer, outputmf, is a fixed output membership function layer and the last is the output. The basic block diagram of ANFIS with input and output is illustrated in Figure 1.
Typical Adaptive neuro-fuzzy Inference System (ANFIS) units.
Figure 1 contains different units of the ANFIS system. Input collected from sensors is trained in various steps. In the first step of fuzzification, raw data is collected, and the fuzzy interface system creates different rules artificially. The created rules are then further trained using the Sugeno method in MATLAB with the help of the artificial neuro-network. In the last step of de-fuzzification, the fuzzified data are again converted into raw variables. The main goal of this entire process is to minimize human effort and overcome manual errors.
There is a need to prioritize the actual parameters of fire so that false alarms could be reduced. To solve this particular problem, a fire monitoring system should be developed to monitor these parameters in real-time and quick action should be taken to reduce fire damage and save human life. The proposed Sugeno-based adaptive neuro-fuzzy interference system decides the presence of fire according to fuzzy rules and vital parameters collected from different sensors.
The designed real-time system collects data like temperature, smoke, humidity, and flame presence to provide the updated status to the owner using the GSM module.
2.2. Architecture of Proposed FDWS
The architecture of the proposed fire detection and warning system is illustrated in Figure 2. The combination of software and hardware together create an automatic fire detection system. The system has different sensors such as a smoke sensor, temperature and humidity sensor, and a flame sensor. These sensors collect data from sensor nodes and then transmit it to the GUI in MATLAB. The data gathered from sensors is then provided as a raw data to fuzzy logic as a linguistic variable which is trained in adaptive neuro-fuzzy system to detect fire status. If the parameters show that the probability of fire is critical, then a message will be sent through GSM regarding the fire condition to the fire controllers and the house owner.
The architecture of the fire alarm system.
The whole system consists of two phases. The first phase of the hardware design includes the development of sensor nodes and the other phase consists of a MATLAB simulation. Both phases are further described in detail.
2.1.1. Hardware Development in Proposed FDWS
In this phase, we designed sensor nodes for fire monitoring employing multi-sensors such as temperature, humidity, smoke, and flame. An Arduino UNO atmega328p micro-controller is used to embed the sensors.
For gathering heat and humidity measurements, DHT22 is used which gives us two important measurements required for a smart fire monitoring system. It gives an output in degrees Celsius for temperature and percentage of humidity. This sensor is shown in Figure 3.
Temperature and humidity sensor used in the proposed system.
The used flame sensor detects the flame at the range of 3 feet and at a 60 degree angle. The LED light shows the presence of fire. Figure 4 represents the flame sensor used in the proposed system.
Flame sensor used in the proposed system.
The MQ-7 Gas Sensor (see Figure 5) is used for the proposed system which is sensitive to carbon monoxide. Carbon monoxide results in a burning process. Its output boots with the concentration of CO level. It Can detect CO anywhere from 20 to 2000 ppm.
Smoke sensor used in the proposed system.
All these sensors are attached to the Arduino UNO ATmega328p board.
2.1.2. PLX-DAQ
After successful configuration with Arduino UNO microcontroller, the PLX-DAQ Macro is used to acquire sensor data from the Arduino UNO to a Microsoft Excel sheet. We just downloaded the PLX-DUQ version 2 and established a connection with Arduino with the simple connection setting like baud rate set at 9600, and port 3 is selected in PLX-DAQ window.
2.1.3. Coding in Arduino IDE
Arduino version 1.8.9 is used to program the hardware configuration of sensors. Coding is done in C language and built in libraries like dht.h for temperature and humidity sensor and MQ7.h for the gas sensor is used. The downloaded code is customized a little bit to get the desired results. The Arduino UNO is a simple way of communication between computer and microcontroller.
The AT MEGA328p connects the UNO serial port, e.g., COM3 with the computer USB port which appears as a virtual COM in the PLX-DAQ. It is simple to use and just need to define connection settings to connect it with an Arduino. The embedded program code in the Arduino UNO board acquires sensor readings and represents it in the PLX-DAQ spreadsheet.
