Locations of sampling points and odor emission from experimental data [9].
\r\n\t• Role of technological innovation and corporate risk management
\r\n\t• Challenges for corporate governance while launching corporate environmental management among emerging economies
\r\n\t• Demonstrating the relationship between environmental risk management and sustainable management
\r\n\t• Contemplating strategic corporate environmental responsibility under the influence of cultural barriers
\r\n\t• Risk management in different countries – the international management dimension
\r\n\t• Global Standardization vs local adaptation of corporate environmental risk management in multinational corporations.
\r\n\t• Is there a transnational approach to environmental risk management?
\r\n\t• Approaches towards Risk management strategies in the short-term and long-term.
In previous studies, Chung and Chiang [6] observed flow patterns and contaminate particle paths in lavatories by using a numerical simulation, and the diameter of the particle is less than 1 mm. They used a floor exhaust ventilation system for analysis, and the target is to improve the air quality of the lavatories’ environment. Their simulation results present only 1.2% and 1.5% of the contaminate particles remain in the men’s and women’s rooms, respectively. With proper ventilation system, the bathroom may have higher efficiency of purging odors in the bathroom.
Sandberg [7],Chung and Hsu [8] defined ventilation efficiency, local air quality index and investigated related impacts of different ventilation patterns arranged by two inlet and outlet diffusers at different locations. In modern society, problem of floor exhaust vent is complex, and tradition ventilation system is mounted a small exhaust fan in the ceiling. Its main function is the suction of moisture air and odors. The air sucked into an exhaust pipe is taken out of house, and then the bathroom becomes negative pressure to let fresh air from outdoors flow into the bathroom.
According to the study of different ventilation patterns, Tung et al. [9] have a novel idea that a mechanical ceiling-supply and wall-exhaust ventilation system remove unpleasant moist air and odors. Although the design is a typical ceiling-exhaust ventilation system, it has a significant difference in the removal of the moisture and odors. The experimental result has showed that doubling the flow rate of supply air from 8.5 to 17 air changes per hour (ACH) leads to a moderate 4% increase in ORE, but a ventilation rate of 8.5 ACH rates is suitable for view of energy-saving. Our study aims to increase ventilation efficiency by numerically investigating different ACH rates and Fire Dynamics Simulator (FDS) [10] is used to analyze the effect of exhaust area. Deployment of local air quality (QI) and odor removal efficiency (ORE) metrics shows great potential to quantitatively evaluate the effectiveness of odor removal in the bathroom [11]. Contaminant source non-uniformity was found to have a great influence on the QI and ORE non-uniformity, with the two tested air distribution methods.
The second case study aims to numerically investigate the operation modes of mechanical ventilation in underground tunnels and compare the effectiveness of CFD with full-scale experiments and existing fire codes in Taiwan. In underground railroad tunnels and subway stations, direct exposure to fire is typically not the most immediate threat to passengers’ lives. Instead, smoke inhalation due extreme heat and toxic gases is the main cause of fatality. For fire safety and hazard mitigation reasons, it is of great importance to investigate the mechanisms thermal-induced smoke motion. On the other hand, the current performance-based fire codes in Taiwan for special structures require full-scale tests [12] to experimentally investigate the effectiveness of the smoke control systems in subway stations before opening for public service. One key feature of fire-induced phenomena is called chimney or stack effect [13], which plays a significant role in the smoke control of subway stations due to various structures of vertical spaces exist such that the buoyancy force of hot smoke will be greatly enhanced.
Due to land scarcity in greater Taipei area, continuously conversion from existing railroad to underground are performed to increase the land utility at different phases and span many years. In addition, integration issues between several transportation systems such as mass transit and high speed rail are also of great concerns. Fourth phase of Nangkang North Tunnel was completed and in full operation in September, 2009. During the construction period, fire codes in Taiwan for special constructions (e.g. underground stations) are also going through various revisions while internationally, new regulations such as AS 4391 in Australia [14] and NFPA130 in USA [15] have been validated and updated. In order to verify the compliance of fire codes nationally and internationally, full scale experiments have been performed before actual operation with emphasis on various operation modes of mechanical ventilation. This paper aims to numerically investigate and compare the effectiveness of CFD with experimental results. Typical road tunnel fire with different causes and kinds of fire were numerically analyzed before and the temperature and visibility of tunnel zone were calculated [16]. The simulation results show that the highest central axis section reaches to 1000°C, and the highest temperature of side wall reaches to 989°C under large fire scales. [17] Furthermore, other tunnel fires are also of great concern due to catastrophic fatality and numerically simulated such as Howard Street Tunnel Fire [17], tunnel fire and tunnel lining failure [18], road tunnels and ventilation effects [19]. On the other hand, smoke control of fires in subway stations and stack effects on smoke propagation are simulated and compared with experiments with great accuracy [20, 21]. Ventilation strategies and application to fire in a long tunnel has been considered [22] and ordinary and emergency ventilations are operated by using dedicated supply and extraction ducts. Another key parameter for tunnel fires is the critical speed and simulation results show a good correlation with real fires [23].
According to current fire codes in Taiwan, a recent version which includes performance-based tests is stipulated in 2008 [12] for the purpose of aiding a complete implementation of fire equipments in underground tunnels and subway stations. In chapter 3 of the fire code, one exception is that simulation study is allowed for the difficult-to-test circumstances, but need to be examined by a special committee on a case-by-case basis. In particular, ventilation is a crucial element while managing fire events in tunnels and control strategy for fire-induced smoke is of great concern. Adoption of numerical simulation a validation tool is probably pioneered in NFPA130 2003, version [15] emergency smoke management session of 7.1.2.4. The first full-scale experiment in Taiwan is the Banciao station in 1999 and the actual tests also laid a foundation of performance-based fire codes for future implementation [24]. In what follows various subway stations [25-27] necessitate the full-scale tests and several reports have been reviewed and published before opening to the public service.
In this section, we briefly present the governing equations of FDS and an outline of the general solution procedure which we used for all the simulations. Details of the individual equations are described in original technical guide [1]. The governing equations are essentially a complete set of partial differential equations with appropriate simplifications and approximations noted. The numerical method consists of a finite difference approximation of the governing equations and a procedure for updating these equations in time [1].
The basic conservation equations for mass, momentum and energy for a Newtonian fluid can be found in almost any textbook on fluid dynamics or CFD. FDS adopts a description of the equations, the notation used, and the various approximations employed in Anderson et al. [28]. Note that this is a set of partial differential equations consisting of six equations for six unknowns, all functions of three spatial dimensions and time: the density r, the three components of velocity u = [u;v;w]T, the temperature T, and the pressure p.
Mass conservation can be expressed either in terms of the density, ρ,
Here
The momentum equation in conservative form is written:
The term uu is a diadic tensor. In matrix notation, with u = [u;v;w]T, the diadic is given by the tensor product of the vectors u and u^T. The term
The term
The energy conservation equation is written in terms of the sensible enthalpy,
The sensible enthalpy is a function of the temperature:
Note the use of the material derivative,
The term
where k is the thermal conductivity.
An approximate form of the Navier-Stokes equations appropriate for low Mach number applications is used in the model.
Here
FDS uses a second-order accurate finite-difference approximation to the governing equations on a series of connected recti-linear meshes. The flow variables are updated in time using an explicit second-order Runge-Kutta scheme. The typical solution algorithm is used as the following to advance in time the density, species mass fractions, velocity components, and background and perturbation pressure. Let
Compute the “average” velocity field
Estimate
Exchange values of
Apply boundary conditions for
Compute the divergence,
Solve the Poisson equation for the pressure fluctuation with a direct solver on each individual mesh:
Estimate the velocity at the next time step
Check the time step at this point to ensure that the stability criteria are satisfied
This concludes the “Predictor” stage of the time step. Next the “corrector” stage follows:
Compute the “average” velocity field
Apply the second part of the Runge-Kutta update to the mass variables. For example, the density is corrected
Exchange values of
Apply boundary conditions for
Compute the divergence
Compute the pressure fluctuation using estimated quantities
Update the velocity via the second part of the Runge-Kutta scheme
At the conclusion of the time step, values of
The details of the predictor-corrector scheme can be found in [1].
We use FDS to build a virtual-model bathroom dimension which is 2.36 m (length), 2.22 m (width), and 2.36 m (height), shown in Fig. 1(a). Bathroom facilities included a bathtub, a sink, and a toilet in the bathroom model. The red, blue, and purple model represents a bathtub, a sink, and a toilet, respectively. In the ceiling, the red square replaces the supply air vent (a mechanical fan), and dimensions are 0.2 m (length) and 0.4 m (width). On the wall, the red square replaces the exhaust air vent (a free opening), and dimensions are 0.2 m and 0.4 (length & height), respectively. The exhaust airflow rate was 1.15 times the supply airflow rate.
Malodorous volatile is frequently manufactured from human feces and urine in bathrooms. The odor source is created on the virtual toilet of bathroom model. This paper follows previous study [9] that assume different distances, which is the distance between the odor source and the wall-exhaust vent. As shown in Fig. 1(b), there are cases of three different distances contained 0.8 m (Case 1), 1.1 m (Case 2), and 2.05 m (Case 3) in this study with ventilation condition of 8.5 Air Change per Hour (ACH).
a) A mock-up bathroom used to analyze the influence of ventilation rates 6.5, 8.5,17, and 24 ACH on the odor removal and shows numerically measured locations of sampling points [9]. (b) In three cases, it reveals related position of bathroom facilities.
According to experiment of the previous study, our study set some issues of four different ACH rates in case 3, and they divide 6.5, 8.5, 17, and 24 ACH rates. Besides, we also change area of wall-exhaust vents contained two cases. One case is 0.24 m (length) and 0.24 m (height), the other is 0.24 m (length) and 0.48 m (height).
By setting a total of 12 concentration sensors, we can gain odor distribution data at specific positions. Odor flow pattern can be obtained and Table 1 shows positions of collected data from P1 to P12. Marked as P and P1–P12 in Fig. 3, ten measuring points (P1–P10) within the bathroom were selected to monitor the concentration levels of odors at these locations, and two extra measuring points were setup outside the room, one (P11) behind the exhaust air vent and another (P12) in front of the supply air vent; moreover, P was the location of a source of odors generated by a person sitting on the toilet in the bathroom.
Sampling location | X-direction (m) | Y-direction (m) | Z-direction (m) |
P1 | |||
Case 1 | 0.30 | 1.78 | 1.16 |
Case 2 | 0.59 | 2.08 | 1.16 |
Case 3 | 1.90 | 1.78 | 1.16 |
P2 | 1.42 | 1.58 | 1.56 |
P3 | 1.42 | 1.58 | 1.16 |
P4 | 0.61 | 1.58 | 1.56 |
P5 | 0.61 | 1.58 | 1.16 |
P6 | 1.42 | 0.79 | 1.56 |
P7 | 1.42 | 0.79 | 1.16 |
P8 | 0.61 | 0.79 | 1.56 |
P9 | 0.61 | 0.79 | 1.16 |
P10 | 1.04 | 1.17 | 1.40 |
P11 | 0.00 | 1.12 | 0.76 |
P12 | 1.11 | 1.18 | 2.36 |
Oder emission | |||
P | |||
Case 1 | 0.50 | 2.07 | 0.40 |
Case 2 | 0.50 | 2.07 | 0.40 |
Case 3 | 1.90 | 1.86 | 0.40 |
Locations of sampling points and odor emission from experimental data [9].
This study followed the same definition and employed a local air quality index to describe the ventilation system’s efficiency in removing odors [9, 29, 30]. A higher score on the local air quality index indicates a better ventilation efficiency due to a lower concentration level. The local air quality index, QI, is defined as follows:
where
Again the same terminology of odor removal efficiency was used to express the ventilation efficiency of the whole bathroom [9, 30, 31]. A lower index indicated a greater difficulty in the removal of odors. The odor removal efficiency, ORE, is defined as follows:
where
As illustrated in Fig. 2(a), the concentration of odors at the 12 observing points (P1–P12) was examined. In these four cases, P1 revealed that odor concentration higher than odor concentration of other pieces points in 6.5 ACH rates, since it’s the shortest distance from toilet. We have found the effect of high temperature led to thermally induced buoyancy such that particle concentration of the higher levels observing points are smaller than the lower level counterparts. Points P2, P4, P6, and P8 were positioned under the ceiling-supply vent and at a relative height of 0.66 of the room height [9]; hence, their concentrations were more than the concentration of points P3, P5, P7, and P9. Numerically, the concentration profile of odors in the cases of 8.5, 17 and 24 ACH was similar to the one in the case of 6.5 ACH and this result is in agreement to experiments [9]. Similar agreement between simulation and experiment is found that higher room ventilation rates resulted in lower absolute concentration levels of odors.
As shown in Fig. 2(b), it was numerically observed that changes in the room ventilation rate affected QI in a significant way at 10 points in the bathroom. All numerically monitored points in the case of 6.5 ACH had a smaller QI than those same points in the cases of 8.5, 17, and 24 ACH. It was noted that the experimentally observed QI values were less than 1.0 at many points with low ventilation rate as compared to numerical values of 1.2. Small discrepancy exist while the same conclusion can be drawn that the ventilation rate of 6.5 ACH is not suitable the bathroom ventilation system. Both numerical and experimental results suggest the cases of 8.5, 17, and 24 ACH had better ventilation efficiency due to QI levels were more than 1.2. Furthermore, a higher ventilation rate resulted in less spreading of odors and improved the local air quality in the bathroom.
Effect of ventilation rates on (a) Concentration profile and (b) QI profile
As illustrated in Fig. 3(a), the concentration of odors at the 12 observing points (P1–P12) was examined. In these four cases of doubling exhaust-vent area, P1 revealed that odor concentration higher than odor concentration of other pieces points in 6.5 ACH rates, since it’s the shortest distance from toilet to there. We have found the effect of high temperature led to thermally induced buoyancy such that particle concentration of the higher levels observing points is smaller than the lower level counterparts. Points P2, P4, P6, and P8 were positioned under the ceiling-supply vent and at a relative height of 0.66 of the room height [9] ; hence, their concentration was more than the concentration of points P3, P5, P7, and P9, which were positioned at a relative height of 0.49. As shown in Fig. 3(b), it was numerically observed that the QI values were less than 1.3 at many points with this low ventilation rate, further improvement from 1.2 when doubling the exhaust-vent area.
Effect of doubling exhaust-vent area on (a) Concentration profile and (b) QI profile.
Effect of locations of a toilet on (a) Concentration profile and (b) QI profile.
The present study numerically investigated three different locations of the toilet in the bathroom and compared with experiments [9]. The distance from the exhaust air vent to the toilet (L) was 0.8, 1.1, and 2.05 m for Cases 1, 2, and 3, respectively. In all three cases, as seen in Fig. 4(a), point P1 reported the highest concentration of odors because this was the nearest location to the source of odors, which is in good agreement with reported data [9]. Similarly, the odor concentration at P1 in Cases 1 and 2 was lower than the concentration in Case 3 due to the fact that toilet in Cases 1 and 2 was in close proximity to the exhaust air vent than in Case 3 and thus had lower odor concentration than Case 3.
Fig. 4(b) shows calculation results that changes in L affected QI at 10 points in the bathroom. It was observed numerically that most observed points in Case 1 had the highest QI, and that their QI levels were more than 0.6 [9]. In summary, Case 1 had the best ventilation efficiency among the three cases, which is in consistent with measured data.
a) Effectiveness analysis of QI at various ACH and comparison with experiments [9]. (b) Effectiveness analysis of ORE at various ACH and comparison with experiments [9].
A curve fitting performed using 3 order polynomial with numerically simulation‘s data. Fig. 5(a) show QI values of P1 in the different ACH rates and the curve fitting these QI values. The curve and point plotted in figure present relation between the different ACH rates and QI, and they illustrate trend that what ACH rates have the best QI value. The light green curve, red curve, and green curve are the experimental data from previous study [9], data in case 3, and data with doubling exhaust-vent area in case 3. The red curve is similar to the light green one, and the trends of both are very near. The Green curve has the translation because the mesh of numerical simulations is not fine enough and turbulent flow partly results in this impact.
Fig. 5(b) shows ORE values in the different ACH rates and the curve fitting ORE values. The curve and point plotted in figure present relation between the different ACH rates and ORE, and they illustrate trend that what ACH rates have the best ORE value. The curve and point plotted in figure present relation between the different ACH rates and QI, and they illustrate trend that what ACH rates have the best QI value. The light green curve, red curve, and green curve are the experimental data from previous study [9], data in case 3, and data with doubling exhaust-vent area in case 3. The red curve and green are similar to the light green one, and the trends are very similar. The ORE value with doubling area of exhaust vent is higher other ORE value in any ACH rates. Doubling area of exhaust vents from 0.0576 m2 to 0.1152 m2 leads to a moderate 30% increase in ORE in 8.5 ACH rates. At the same condition, it leads to a moderate 16% increase in ORE in 17 ACH rates.
Smoke accumulation and migration are two main fire hazards in tunnel fires. The worst case scenario happens when the mechanical ventilation fails. Therefore, how to design a fail-safe system in underground station is of great importance. Typically, effective egress time is calculated as the smoke layer descends till 1.8 meter above the ground. In this study, we simultaneously compare the experiments and CFD simulations regarding two different fire loads of 1 and 5MW for 6 minutes and 20 minutes, respectively.
Fire dynamics simulator, FDS 6.0 [10] is a computational fluid dynamics model of fire-driven fluid flow. The software solves numerically a tform of the Navier-Stokes equations appropriate for low-speed, thermally-driven flow, with an emphasis on smoke and heat transport from fires. Figure 6 shows the schematics of underground railroad station constructed using FDS software. It is noted that two underground levels, B1 and B2, are simulated based on actual dimensions. Moreover, two rectangular openings in 4m*5m*0.3m shown in Fig.1 are used to represent the escalator positions between two underground levels. Total dimensions of 16m*130m*10m space are simulated with a total of more than 3 million computational grids in FDS model. Underground station is shown in blue colour with dimension of 8m*130m*1m while the wall thickness between B1 and B2 is 30cm (as shown in red). Experimentally, two different fire loads of 1 and 5 MW are ignited at the centre of underground station using the standard methods [14, 15] and the smoke descending positions are measured across the station area at 6 and 20 minutes, respectively.
Schematics of underground railroad station constructed using FDS software. Two underground levels, B1 and B2, are simulated and two rectangular openings are used to represent the escalator positions between two underground levels.
Numerically, smoke patterns under fire load of 1 and 5 MW at time span of 360 and 1200 seconds are shown in Fig 7-8, respectively. It is noted that all mechanical ventilations are set to be in-operation to simulate the worst case scenario of mechanical failures. The particle function in FDS is used as a representation of smoke patterns. According to the fire code of NFPA130, 4 minutes of egression time is required and the key is to avoid smoke descending below 1.8m. For both fire loads at 1 and 5 MW, we can observe that smoke particles occupy less than 5% in area in terms of 1.8m height, especially in the escalator positions. It is considered that the simulation results in compliance with NFPA130 and agrees well with full scale experiments. Accordingly, 20 minutes of smoke accumulation and migration due to forced convection for both 1 and 5 MW fires will result in 50% and 99% smoke layers below 1.8m in escalator areas, which are considered hazardous for egression and unable to escape.
We can summarize the simulation results and compare with previous experiments for 1MW fire load after 6 minutes in table 2. Generally speaking, simulations results are in a good agreement with experimental observations. For the experiment, only visual observations by the inspectors may not be able to detect accumulation of smoke layer, especially when the smoke in little in volume and transparency optically. However, computer simulations enable us to numerically visualize the smoke at any location and scientifically calculate any vital information required. In this case, smoke layer height is of great importance due to egress needs, we can calculate tiny smoke accumulation in the areas located at as front and back station. Numerically, we can find less than 5% in area for smoke height<1.8m at both front and back station and the reason is because the escalator/stair induced chimney effect. Overall, both simulation and experiment are in good agreement and according to the fire code of NFPA130, 4 minutes of egression time is satisfied with 50% allowance time before smoke descending below 1.8m.
Smoke patterns under fire of 1MW at (a) 360s and (b) 1200s, respectively
Smoke patterns under fire of 5MW at (a) 360s and (b) 1200s, respectively
Location Area | Front (5m from walls) | Center | Back (5m from walls) | |
Simulation | Tunnel | <5% in area for smoke height<1.8m | Higher than 1.8m | Higher than 1.8m |
Station | <5% in area for smoke height<1.8m | Higher than 1.8m | <5% in area for smoke height<1.8m | |
Experiment [27] | Tunnel | Higher than 1.8m | Higher than 1.8m | Higher than 1.8m |
Station | Higher than 1.8m | Higher than 1.8m | Higher than 1.8m |
Comparison of simulation results and experimental observations.
Two mechanical fans located on both sides of underground station B2 as indicated in green colour. Fire load is individually 1MW and the time is 1~6 minutes. Both front (left) and parametric (right) views are presented for comparison of smoke management effects.
Two mechanical fans located on both sides of underground station B2 as indicated in green colour. Fire load is individually 5MW and the time is 1~6 minutes. Both front (left) and parametric (right) views are presented for comparison of smoke management effects.
Two mechanical fans (volume flow rate at 525m3/sec) located on both sides of underground station B2 as indicated in green colour in Fig.9-10. Fire load are 1MW and 5MW and the time is 1~6 minutes. It is simulated that smoke can be greatly diffused due to fan-driven flows, resulting in less than 1% area of 1.8m smoke height and relatively safe egress mode after 6 minutes of 1MW fire. Simulation results agree with full scale observations [27] where located fans near escalator area would effectively re-distribute the thermally-induced smoke to the overall ceilings before accumulating in the escape route.
From the simulations on mechanical ventilation of previous sections, the following conclusions can be drawn:
The first case numerically studies the effect of ventilation rates and bathroom equipment locations on the odor removal efficiency in a modeled bathroom. Our study agrees well with the results of a previously published study [9].
Qualitatively, numerical calculations result in similar trend such that the influence of ventilation rates and toilet locations on the odor removal is consistent with test data in a mock-up bathroom. Similar conclusion is drawn that higher ventilation rates and shorter distances between the toilet and exhaust air vent are found to be more capable of raising local air quality and the removal efficiency of odors.
Quantitatively, two indexes of QI and ORE, are employed for the analysis of the ventilation effect. The experimental results indicate that doubling the flow rate of supply air from 8.5 to 17 ACH leads to a moderate 4% increase in ORE, while the result of our numerical study is found that doubling area of exhaust vents leads to a moderate 30% increase in ORE in 8.5 ACH rates. That is, if we want design bathroom from the point of view’s energy-saving, the new idea is that increasing exhaust-vent area also arise more efficient of ventilation system than enhancing ACH rates.
The second case study aims to numerically investigate the operation modes of mechanical ventilation in underground tunnels and compare the CFD effectiveness with full-scale experiments and existing fire codes in Taiwan. We numerically investigate the effectiveness of the smoke control systems via mechanical ventilation. In particular, focus is placed on safe egress time in different operation modes and comparison with full scale experiments as well as fire codes in Taiwan. The following conclusions can be drawn from simulation results.
Fire scenarios are very complex phenomena and these are especially true for the cases of underground channels and subway stations. It is simulated that for the case of Nangkang North Tunnel in Taipei area, the worst case scenario of mechanical failures in ventilation systems will still hold the egress time of six minutes or more under fire loads of 1 and 5MW, respectively. The conservative design criterion is in compliance with NFPA 130 and domestic fire codes in Taiwan. Moreover, it is imperative to perform full-scale experiment for special structures such as underground station, i.e., performance-based fire codes and our simulation results agree well experiments.
Current design of underground station in Taiwan relies heavily on the smoke storage space to extend the egress time while the mechanical ventilation systems have only a minor contribution in smoke management. On the other hand, mechanical ventilation systems can play a more significant role in rescue mission after successful personnel evacuation. For example, numerical results indicate that smoke chimney effect can be effectively dispersed through mechanical ventilation to both safeguard firefighters and rescue members.
In terms of smoke control strategy, transient simulations are imperative in order to test the fire codes of various parametric values and different scenarios. Transient simulations allow one to check not only the time required to confine the smoke front, but also the extension of all the surrounding environment which can be potentially occupied by smoke. Educationally, it is also a very good teaching tool to train the firefighters and managers in order to construct a better rescue strategy.
Thanks for the National Institute of Standards and Technology develop Fire Dynamics Simulator and Smokeview (FDS-SMV).
Kalman filtering is an algorithm that provides estimates of some unknown variables given the measurements observed over time. Kalman filters have been demonstrating its usefulness in various applications. Kalman filters have relatively simple form and require small computational power. However, it is still not easy for people who are not familiar with estimation theory to understand and implement the Kalman filters. Whereas there exist some excellent literatures such as [1] addressing derivation and theory behind the Kalman filter, this chapter focuses on a more practical perspective.
\nFollowing two chapters will devote to introduce algorithms of Kalman filter and extended Kalman filter, respectively, including their applications. With linear models with additive Gaussian noises, the Kalman filter provides optimal estimates. Navigation with a global navigation satellite system (GNSS) will be provided as an implementation example of the Kalman filter. The extended Kalman filter is utilized for nonlinear problems like bearing-angle target tracking and terrain-referenced navigation (TRN). How to implement the filtering algorithms for such applications will be presented in detail.
\nKalman filters are used to estimate states based on linear dynamical systems in state space format. The process model defines the evolution of the state from time \n
where \n
The process model is paired with the measurement model that describes the relationship between the state and the measurement at the current time step \n
where \n
The role of the Kalman filter is to provide estimate of \n
Kalman filter algorithm consists of two stages: prediction and update. Note that the terms “prediction” and “update” are often called “propagation” and “correction,” respectively, in different literature. The Kalman filter algorithm is summarized as follows:
\nPrediction:
\nPredicted state estimate | \n\n\n | \n
Predicted error covariance | \n\n\n | \n
Update:
\nMeasurement residual | \n\n\n | \n
Kalman gain | \n\n\n | \n
Updated state estimate | \n\n\n | \n
Updated error covariance | \n\n\n | \n
In the above equations, the hat operator, \n
The predicted state estimate is evolved from the updated previous updated state estimate. The new term \n
In the update stage, the measurement residual \n
We need an initialization stage to implement the Kalman filter. As initial values, we need the initial guess of state estimate, \n
Note that Kalman filters are derived based on the assumption that the process and measurement models are linear, i.e., they can be expressed with the matrices \n
An example for implementing the Kalman filter is navigation where the vehicle state, position, and velocity are estimated by using sensor output from an inertial measurement unit (IMU) and a global navigation satellite system (GNSS) receiver. In this example, we consider only position and velocity, omitting attitude information. The three-dimensional position and velocity comprise the state vector:
where \n
where \n
where \n
Now, we have the process model as:
where
The GNSS receiver provides position and velocity measurements corrupted by measurement noise \n
It is straightforward to derive the measurement model as:
where
In order to conduct a simulation to see how it works, let us consider \n
We need to generate noise of acceleration output and GNSS measurements for every time step. Suppose the acceleration output, GNSS position, and GNSS velocity are corrupted with noise with variances of 0.32, 32, and 0.032, respectively. For each axis, one can use MATLAB function randn or normrnd for generating the Gaussian noise.
\nThe process noise covariance matrix, \n
Let us start filtering with the initial guesses
where \n
In this simulation, \n
The time history of estimation errors of two Monte-Carlo runs is depicted in Figure 1. We observe that the estimation results of different simulation runs are different even if the initial guess for the state estimate is the same. You can also run the Monte-Carlo simulation with different initial guesses (sampled from a distribution) for the state estimate.
\nTime history of estimation errors.
The standard deviation of the estimation errors and the estimated standard deviation for x-axis position and velocity are drawn in Figure 2. The standard deviation of the estimation error, or the root mean square error (RMSE), can be obtained by computing standard deviation of \n
Actual and estimated standard deviation for x-axis estimate errors.
In real applications, you will be able to acquire only the estimated covariance because you will hardly have a chance to conduct Monte-Carlo runs. Also, getting a good estimate of \n
Source code of MATLAB implementation for this example can be found in [5]. It is recommended for the readers to change the parameters and aircraft trajectory by yourself and see what happens.
\nSuppose you have a nonlinear dynamic system where you are not able to define either the process model or measurement model with multiplication of vectors and matrices as in (1) and (2). The extended Kalman filter provides us a tool for dealing with such nonlinear models in an efficient way. Since it is computationally cheaper than other nonlinear filtering methods such as point-mass filters and particle filters, the extended Kalman filter has been used in various real-time applications like navigation systems.
\nThe extended Kalman filter can be viewed as a nonlinear version of the Kalman filter that linearized the models about a current estimate. Suppose we have the following models for state transition and measurement
where \n
All you need is to obtain the Jacobian matrix, first-order partial derivative of a vector function with respect to a vector, of each model in each time step as:
Note the subscripts of \n
Prediction:
\nPredicted state estimate | \n\n\n | \n
Predicted error covariance | \n\n\n | \n
Update:
\nMeasurement residual | \n\n\n | \n
Kalman gain | \n\n\n | \n
Updated state estimate | \n\n\n | \n
Updated error covariance | \n\n\n | \n
As in the Kalman filter algorithm, the hat operator, \n
We are going to estimate a 3-dimensional target state (position and velocity) by using measurements provided by a range sensor and an angle sensor. For example, a radar system can provide range and angle measurement and a combination of a camera and a rangefinder can do the same. We define the target state as:
where \n
The process noise has the covariance of \n
The measurement vector is composed of line-of-sight angles to the target, \n
where \n
where \n
In the simulation, the sensor is initially located at \n
Trajectory of the sensor and the target.
In the filter side, the covariance matrix for the process noise can be set as:
where \n
\n\n
The above equation means that the error of the initial guess for the target state is randomly sampled from a Gaussian distribution with a standard deviation of \n
Time history of an estimation result for x-axis position and velocity is drawn together with the true value in Figure 4. The shape of the line will be different at each run. The statistical result can be shown as Figure 5. Note that the filter worked inconsistently with the estimated error covariance different from the actual value. This is because the process error covariance is set to a very large number. In this example, the large process error covariance is the only choice a user can make because the measurement cannot correct the velocity. One can notice that the measurement Eq. (26) has no term dependent on the velocity, and therefore, matrix \n
Time history of an estimation result for x-axis position and velocity.
Actual and estimated standard deviation for x axis estimate errors.
Source code of MATLAB implementation for this example can be found in [5]. It is recommended for the readers to change the parameters and trajectories by yourself and see what happens.
\nTerrain-referenced navigation (TRN), also known as terrain-aided navigation (TAN), provides positioning data by comparing terrain measurements with a digital elevation model (DEM) stored on an on-board computer of an aircraft. The TRN algorithm blends a navigational solution from an inertial navigation system (INS) with the measured terrain profile underneath the aircraft. Terrain measurements have generally been obtained by using radar altimeters. TRN systems using cameras [7], airborne laser sensors [8], and interferometric radar altimeters [9] have also been addressed. Unlike GNSS’s, TRN systems are resistant to electronic jamming and interference, and are able to operate in a wide range of weather conditions. Thus, TRN systems are expected to be alternative/supplement systems to GNSS’s.
\nThe movement of the aircraft is modeled by the following Markov process:
where \n
Conventional TRN structure.
Typical TRN systems utilize measurements of the terrain elevation underneath an aircraft. The terrain elevation measurement is modeled as:
where \n
Relationship between measurements in TRN.
The process model in (33) and the measurement model in (34) can be linearized as:
where \n
The DEMs are essentially provided as matrices containing grid-spaced elevation data. For obtaining finer-resolution data, interpolation techniques are often used to estimate the unknown value in between the grid points. One of the simplest methods is linear interpolation. Linear interpolation is quick and easy, but it is not very precise. A generalization of linear interpolation is polynomial interpolation. Polynomial interpolation expresses data points as higher degree polynomial. Polynomial interpolation overcomes most of the problems of linear interpolation. However, calculating the interpolating polynomial is computationally expensive. Furthermore, the shape of the resulting curve may be different to what is known about the data, especially for very high or low values of the independent variable. These disadvantages can be resolved by using spline interpolation. Spline interpolation uses low-degree polynomials in each of the data intervals and let the polynomial pieces fit smoothly together. That is, its second derivative is zero at the grid points (see [11] for more details). Classical approach to use polynomials of degree 3 is called cubic spline. Because the elevation data are contained in a two-dimensional array, bilinear or bicubic interpolation are generally used. Interpolation for two-dimensional gridded data can be realized by interp2 function in MATLAB. Cubic spline interpolation is used in this example.
\nThe DEM we are using in this example has a \n
Contour representation of terrain profile.
An aircraft, initially located at \n
The process noise \n
Let us consider \n
The above equation means the error of the initial guess for the target state is randomly sampled from a Gaussian distribution with a standard deviation of \n
The time history of RMSE of the navigation is shown in Figure 9. One can observe the RMSE converges relatively slower than other examples. Because the TRN estimates 2D position by using the height measurements, it often lacks information on the vehicle state. Moreover, note that the extended Kalman filter linearizes the terrain model and deals with the slope that is effective locally. If the gradient of the terrain is zero, the measurement matrix \n
Time history of RMSE.
Source code of MATLAB implementation for this example can be found in [5]. It is recommended for the readers to change the parameters and aircraft trajectory by yourself and see what happens.
\nIn this chapter, we introduced the Kalman filter and extended Kalman filter algorithms. INS/GNSS navigation, target tracking, and terrain-referenced navigation were provided as examples for reader’s better understanding of practical usage of the Kalman filters. This chapter will become a prerequisite for other contents in the book for those who do not have a strong background in estimation theory.
\nThe Open Access model is applied to all of our publications and is designed to eliminate subscriptions and pay-per-view fees. This approach ensures free, immediate access to full text versions of your research.
",metaTitle:"Open Access Publishing Fees",metaDescription:"Open Access Publishing Fees",metaKeywords:null,canonicalURL:"/page/OA-publishing-fees",contentRaw:'[{"type":"htmlEditorComponent","content":"As a gold Open Access publisher, an Open Access Publishing Fee is payable on acceptance following peer review of the manuscript. In return, we provide high quality publishing services and exclusive benefits for all contributors. IntechOpen is the trusted publishing partner of over 118,000 international scientists and researchers.
\\n\\nThe Open Access Publishing Fee (OAPF) is payable only after your full chapter, monograph or Compacts monograph is accepted for publication.
\\n\\nOAPF Publishing Options
\\n\\n*These prices do not include Value-Added Tax (VAT). Residents of European Union countries need to add VAT based on the specific rate in their country of residence. Institutions and companies registered as VAT taxable entities in their own EU member state will not pay VAT as long as provision of the VAT registration number is made during the application process. This is made possible by the EU reverse charge method.
\\n\\nServices included are:
\\n\\nSee our full list of services here.
\\n\\nWhat isn't covered by the Open Access Publishing Fee?
\\n\\nIf your manuscript:
\\n\\nYour Author Service Manager will inform you of any items not covered by the OAPF and provide exact information regarding those additional costs before proceeding.
\\n\\nOpen Access Funding
\\n\\nTo explore funding opportunities and learn more about how you can finance your IntechOpen publication, go to our Open Access Funding page. IntechOpen offers expert assistance to all of its Authors. We can support you in approaching funding bodies and institutions in relation to publishing fees by providing information about compliance with the Open Access policies of your funder or institution. We can also assist with communicating the benefits of Open Access in order to support and strengthen your funding request and provide personal guidance through your application process. You can contact us at oapf@intechopen.com for further details or assistance.
\\n\\nFor Authors who are still unable to obtain funding from their institutions or research funding bodies for individual projects, IntechOpen does offer the possibility of applying for a Waiver to offset some or all processing feed. Details regarding our Waiver Policy can be found here.
\\n\\nAdded Value of Publishing with IntechOpen
\\n\\nChoosing to publish with IntechOpen ensures the following benefits:
\\n\\nBenefits of Publishing with IntechOpen
\\n\\nAs a gold Open Access publisher, an Open Access Publishing Fee is payable on acceptance following peer review of the manuscript. In return, we provide high quality publishing services and exclusive benefits for all contributors. IntechOpen is the trusted publishing partner of over 118,000 international scientists and researchers.
\n\nThe Open Access Publishing Fee (OAPF) is payable only after your full chapter, monograph or Compacts monograph is accepted for publication.
\n\nOAPF Publishing Options
\n\n*These prices do not include Value-Added Tax (VAT). Residents of European Union countries need to add VAT based on the specific rate in their country of residence. Institutions and companies registered as VAT taxable entities in their own EU member state will not pay VAT as long as provision of the VAT registration number is made during the application process. This is made possible by the EU reverse charge method.
\n\nServices included are:
\n\nSee our full list of services here.
\n\nWhat isn't covered by the Open Access Publishing Fee?
\n\nIf your manuscript:
\n\nYour Author Service Manager will inform you of any items not covered by the OAPF and provide exact information regarding those additional costs before proceeding.
\n\nOpen Access Funding
\n\nTo explore funding opportunities and learn more about how you can finance your IntechOpen publication, go to our Open Access Funding page. IntechOpen offers expert assistance to all of its Authors. We can support you in approaching funding bodies and institutions in relation to publishing fees by providing information about compliance with the Open Access policies of your funder or institution. We can also assist with communicating the benefits of Open Access in order to support and strengthen your funding request and provide personal guidance through your application process. You can contact us at oapf@intechopen.com for further details or assistance.
\n\nFor Authors who are still unable to obtain funding from their institutions or research funding bodies for individual projects, IntechOpen does offer the possibility of applying for a Waiver to offset some or all processing feed. Details regarding our Waiver Policy can be found here.
\n\nAdded Value of Publishing with IntechOpen
\n\nChoosing to publish with IntechOpen ensures the following benefits:
\n\nBenefits of Publishing with IntechOpen
\n\n