CN109738365B

CN109738365B - Gas leakage source searching method based on TDLAS sensor

Info

Publication number: CN109738365B
Application number: CN201811601199.8A
Authority: CN
Inventors: 孟庆浩; 戴旭阳; 靳荔成
Original assignee: Tianjin University
Current assignee: Tianjin University
Priority date: 2018-12-26
Filing date: 2018-12-26
Publication date: 2021-10-01
Anticipated expiration: 2038-12-26
Also published as: CN109738365A

Abstract

The invention relates to a TDLAS sensor-based gas leakage source searching method, which performs dynamic planning of a sensor light path in a way that a ground mobile robot is matched with a holder, and comprises the following steps: [1]Initializing map and environment parameters; [2]Collecting smell packet capturing times k and wind vector at current position r

Information, updating posterior probability by Bayesian inference; [3]Calculating the global map information entropy value at the current moment, if the entropy value is lower than a certain threshold value, enabling the sensor light path to perform 360-degree scanning by taking the maximum probability point as the center and a certain length as the radius, and performing source confirmation by calculating the smell flux on a closed curve enclosed by the scanning light path; if the entropy value is still higher than the set threshold value, continuing the search path planning at the next moment; [4]And planning the optical path of the sensor at the next moment.

Description

Gas leakage source searching method based on TDLAS sensor

Technical Field

The invention relates to a robot autonomous leakage source searching algorithm based on a TDLAS gas concentration sensor, which realizes accurate positioning of an odor source in an information trend searching mode with additional geometric constraint.

Background

With the continuous development of industrialization, some factories storing high-risk toxic substances frequently generate malignant gas leakage accidents caused by careless supervision. Once leakage occurs, peculiar smell disturbs people and causes panic, and serious loss of life and property safety of people is caused. Therefore, under the condition that toxic or harmful gas leaks, the position of a leakage source is searched quickly and accurately, and the later reasonable treatment and rescue are very important.

The robot is used for carrying various sensors, and can replace workers to enter a dangerous area for leakage investigation. Considering the internal mechanism of bionics smell tracing, when the robot carries out dynamic path planning according to sensor data, the data of historical time needs to be reasonably utilized, rather than only carrying out one-sided decision according to the data of current time. Aiming at the problem, Vergassola and the like propose an information tendency (Informataxis) strategy for maximizing global information gain, and the algorithm effectively fuses current information and historical information through Bayesian inference; meanwhile, the method does not depend on a turbulent diffusion model, and can effectively overcome the problem of source search under a turbulent environment (Vergasssola M, Villereux E, Shraiman B I. 'Infotaxis' as a Strategy for Searching with gradient [ J ]. Nature,2007,445(7126): 406.). The nature of the algorithm to maximize the amount of map information makes the robot more inclined to collect unknown area information at the beginning of the search, but this search pattern takes longer for a robot carrying a single point touch sensor to search for a coverage area without smoke plumes. Meanwhile, in the vicinity of the gas source or in a region with high local concentration, the robot is easy to have a local self-sinking problem (namely, the robot is repeatedly turned in the region), and further the source positioning accuracy is reduced.

TDLAS (tunable semiconductor laser absorption spectroscopy) is a novel optical gas concentration sensor with high selectivity, the output value of the sensor is the sum of all gas concentrations on an optical path, and therefore the linear integration type sensor has incomparable advantages over a single-point contact type sensor in a monitoring range. The Lilienthal team uses a ground robot to carry TDLAS Sensors to reconstruct the dense gas distribution { Arain M A, Trincavelli M, Cirillo M, et al. Global coverage measurement sequences for mobile roboes employing mobile gate with a remote gate sensor [ J ]. Sensors,2015,15(3):6845-6871 }. However, since the gas concentration distribution reconstruction task has a high demand for the sampled data (i.e. the more sampled data, the more accurate the reconstruction result is), the objective of the robot path planning in this study is to maximize the spatial coverage, which is obviously not an efficient search mode, and neither the advantages of the TDLAS sensor in the measurement range nor the important information such as wind speed and direction can be fully utilized.

In summary, the current gas leakage source search algorithm is limited by the measurement range of the single-point contact sensor, but lacks effective utilization of the TDLAS sensor.

Disclosure of Invention

The invention aims to provide a TDLAS gas concentration sensor-based gas leakage source searching method, which not only can give full play to the advantage of large space coverage of a TDLAS sensor, but also can effectively fuse information of other sensors (such as an anemograph), thereby improving the searching efficiency and overcoming local self-trapping. The technical scheme is as follows:

a TDLAS sensor-based gas leakage source searching method performs dynamic planning on a sensor light path in a mode that a ground mobile robot is matched with a holder, and comprises the following steps:

[1]initializing map and environment parameters; setting the probability P that each grid of the area to be searched is judged as the odor source at the initial moment₀(r); initializing parameters such as a source release rate R, a gas diffusivity rate D and the like;

[2]collecting smell packet capturing times k and wind vector at current position r

The information is updated by Bayesian inference, the posterior probability is updated, and the current gas diffusion source is predicted in the detection light path of the sensor

In the above, the concentration value detectable at each point in the grid map

Secondly, the prediction concentration value is utilized to solve the Poisson distribution coefficient of the likelihood function in the Bayesian inference model

Wherein dt is the time interval, q is the average length of the smell packet from the source to the disappearance, and a is the effective search radius of the sensor; finally, after standardized similarity processing, the posterior probability distribution P of the current moment is updated_t(r)；

[3]Calculating the global map information entropy value at the current moment: s_t＝∑P_t(r)·lnP_t(r), if the entropy is lower than a certain threshold, making the light path of the sensor scan 360 degrees by taking the maximum probability point as the center and a certain length as the radius of r, and calculating the scanThe source confirmation is carried out on the smell flux on a closed curve enclosed by the light path, so that the self-trapping problem caused by over-high local concentration is solved; if the entropy value is still higher than the set threshold value, continuing the search path planning at the next moment;

[4] planning a sensor light path at the next moment; determining undetermined detection circuits of the sensors according to undetermined target points, estimating posterior probability distribution and corresponding entropy values of the circuits to be detected of each sensor moved by the optical path according to the probability distribution at the current moment, and selecting one circuit with the largest information entropy reduction as an undetermined detection area of the optical path of the sensor at the next moment;

[5] the robot moves the light path of the sensor to a specified area through self movement and rotation of the holder, and starts iterative computation at the next moment.

The main advantages and the characteristics of the invention are embodied in the following aspects:

1. the invention gives full play to the advantages of the TDLAS gas concentration sensor in the detection range, effectively reduces the exploration time of the robot to the coverage area without smoke plume by matching with the rotation of the holder, and improves the search efficiency.

2. The invention judges whether a gas diffusion source is in the closed curve or not by calculating and scanning the odor packet flux on the closed curve, thereby effectively avoiding the robot from being trapped in local convergence near the gas source or in a region with higher concentration.

Drawings

FIG. 1 is a flowchart of the overall algorithm of the present invention.

Fig. 2 shows a rasterized map at different times, the shades of color representing the magnitude of the probability that the grid is the source of the scent.

Fig. 3 is a geometric constraint newly added to the algorithm, namely that the detection range of the sensor becomes a plurality of points on the same straight line.

Figure 4 shows that source identification can be made by the difference in the flux of scent packets by the scent source within or outside the closed curve.

Fig. 5 is a schematic diagram of optical path planning at the next moment of the sensor, wherein two end points of the optical path have 5 candidates respectively.

Detailed Description

The invention is described in detail below with reference to the figures and examples. The embodiments are specific implementations on the premise of the technical scheme of the invention, and detailed implementation modes and processes are given. The scope of protection of the claims of the present application is not limited by the description of the embodiments below. The algorithm flow is shown in fig. 1, and the steps are as follows:

[1] map and environmental parameters are initialized. Discretizing an area to be searched into an M multiplied by N grid map, wherein each grid represents the probability of being judged as an odor source and meets the uniform distribution at the initial moment, namely the probability that any grid is the odor source is 1/(M multiplied by N); initializing a source release rate R, a gas diffusivity rate D, a robot size a and an odor packet life tau, wherein the partial parameters are used for constructing a likelihood function in subsequent Bayesian inference.

[2]Collecting smell packet capturing times k and wind vectors of a robot at a current position r

And updating the probability that each grid is judged as the odor source by Bayesian inference. Taking fig. 2 as an example, assuming that the probability distribution of the map at the previous time is shown in the left graph, after a bayesian inference update, the probability distribution will change as shown in the right graph.

First, the concentration value c detectable at each point in the grid map when the gas diffusion source is on the sensor detection light path is estimated. Where geometric constraints are introduced, line segments may be introduced as shown in FIG. 3

The upper concentration line integral is broken down into the integrals of L equidistant points, L being 4 in the figure, and (3,1), (4,2), (5,3), (6, 4). On line segment

Any point on the upper line segment and the lower line segment under the condition of existence of the gas source

Subject to a uniform distribution of the probability of the gas source, i.e.

The line source predicted concentration profile can therefore be expressed as a weighted sum of the predicted concentrations for each point source:

wherein c (r | r)₀) The Bessel function K can be obtained by solving the convection diffusion equation₀The analytical formula (2).

Second, assuming that the probability of capturing k flavor packs at location r obeys a poisson distribution (i.e., likelihood function in bayesian inference model) over a finite time interval dt, the concentration distribution is predicted from the line source

The Poisson distribution coefficient in the time period can be calculated

And finally, updating the posterior probability distribution of the current moment by taking the posterior probability of the previous moment as the prior probability of the current moment, and carrying out normalization treatment:

[3]calculating the global map information entropy value at the current moment:

if the entropy value is higher than the set threshold value, continuing the search path planning at the next moment; if the entropy is lower than a certain threshold, as shown in fig. 4, the light path of the sensor is scanned 360 degrees with the maximum probability point as the center and a certain length as the radius, and the odor packet on the closed curve l enclosed by the scanning light path is calculatedFlux.

For a certain infinitesimal element on a closed curve

The direction of which is specified to point along the radius outside the closed curve, the number of detected smell packets in the infinitesimal is k_lThe direction of the wind vector is taken as the moving direction of the smell bag

The odor flux on the closed curve l is then:

the left panel in fig. 4 shows that the flux should be more positive when the gas source is inside the closed curve; the right graph in fig. 4 shows that the flux should be as close to zero as possible when the gas source is outside the closed curve.

[4] And planning the optical path of the sensor at the next moment. Estimating posterior probability distribution of two end points of the light path moving to 5 undetermined target points (upper, lower, left, right and original places) and corresponding entropy values thereof by using the probability distribution at the current moment, and selecting a line with the largest information entropy reduction as an undetermined detection area of the light path of the sensor at the next moment. Since there are 5 candidates at each end point (as shown in fig. 5), there are 25 optical path candidates in total, i.e., 5 × 5 optical path candidates.

Claims

1. A TDLAS sensor-based gas leakage source searching method performs dynamic planning on a sensor light path in a mode that a ground mobile robot is matched with a holder, and comprises the following steps:

[1]initializing map and environment parameters; setting the probability P that each grid of the area to be searched is judged as the odor source at the initial moment₀(r); initializing parameters of a source release rate R and a gas diffusivity rate D;

In the above, the concentration value detectable at each point in the grid map

[3]Calculating the global map information entropy value at the current moment:

if the entropy value is lower than a certain threshold value, the light path of the sensor is scanned for 360 degrees by taking the maximum probability point as the center and a certain length as a radius r, and the source confirmation is carried out by calculating the smell flux on a closed curve enclosed by the scanning light path, so that the self-trapping problem caused by overhigh local concentration is solved; if the entropy value is still higher than the set threshold value, continuing the search path planning at the next moment;