KR101652099B1

KR101652099B1 - Risk map based on gas accident response and prevention system

Info

Publication number: KR101652099B1
Application number: KR1020150179226A
Authority: KR
Inventors: 오정석; 김정훈; 조영도
Original assignee: 한국가스안전공사
Priority date: 2015-12-15
Filing date: 2015-12-15
Publication date: 2016-08-29

Abstract

According to the present invention, a system for handling and preventing an accident based on a risk map comprises: an accident handling module to calculate a static risk of a gas industry facility and a dynamic risk of a worker to provide a handling risk indicating a risk of a risk map for handling an accident; an accident prevention module to consider a degradation model related to a risk and an accident and a failure rate to construct an accident sign prediction model, and provide a prevention risk for predicting a future risk by the accident sign prediction model; a database unit to store accident history information, material information, worker characteristic information, and surrounding environment information; a data warehouse to collect a plurality of risk factors from the database unit, and provide a reconstructed information structure for direct use by the accident handling module and the accident prevention module without separate data processing work; and a risk map-linked drive module linked to the accident handling module and the accident prevention module to combine a probability of accident prediction with an accident handling risk to provide a final risk.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a risk map-

The present invention relates to a risk-based accident response system and an accident prevention system, and more particularly, to a risk-based accident response system and an accident prevention system, And provides a risk map based incident response and accident prevention system that can predict future risks for future accident prevention.

In general, industrial plants and factories are equipped with instrumentation in major gas facilities for risk and safety management, and are monitoring real-time monitoring parameters at the central control center.

In particular, simple analytical tasks that measure, analyze and review risk factors individually, such as pressure, temperature, and vibration, are mainstream in the business world. However, many risk factors are input in parallel, Research on the method is also being carried out.

However, a plurality of risk factor comprehensive analysis methods according to the prior art have the following problems.

First, the prior art expresses the risk through a plurality of risk factors in a specific space or region, so that the accuracy of the risk and accident analysis is improved rather than the simple analysis technique. However, since only the static risk is calculated, It has been difficult to calculate the amount of water.

Second, in using the intelligent algorithm, the prior art has a problem in that the accuracy of the intelligent model is lowered because the algorithm is applied without modification through validation of the domain to be applied.

Third, the conventional technology has a problem that it is difficult to utilize as a preventive system against future risks and accidents due to the use of an accident response-oriented procedure and a system related thereto.

Fourth, the conventional technology has a problem in that it can not provide a single user interface for carrying out overall safety management by being combined with accident response and accident prevention.

Korean Patent Publication No. 10-2015-0089313

The technical problem to be solved by the present invention is to provide a system and a method for providing an accident risk management system in which the present risk for an accident response is differently provided according to a predefined area, And to provide a risk map based incident response and accident prevention system that can predict future risk.

According to an aspect of the present invention, there is provided a risk-based accident response and accident prevention system, comprising: calculating a static risk level of a gas industry facility and a dynamic risk level of a worker, An incident response module; An accident prevention module that establishes an accident predicting model in consideration of a risk and an accident related deterioration model and a failure rate and provides a risk of preventing a future risk prediction through the accident predicting model; A database unit storing accident history information, material information, operator characteristic information, and surrounding environment information; A data warehouse for collecting a plurality of risk factors from the database unit and providing a reconstructed information structure for use directly in the incident response module and the accident prevention module without any separate data processing operation; And a risk map linkage driving module, which is associated with the incident response module and the accident prevention module, and combines the risk of an accident response with the probability of an accident prediction to provide a final risk.

The present invention can build a risk information through various measurement data and efficiently cope with the risk, effectively predict the risk and recognize the failure time of the gas facility in advance, and perform the predictive maintenance to reduce the failure rate and increase the utilization rate In addition to this, it is possible to use the information that the danger level of the map changes periodically according to the movement of the worker, so as to prevent the unauthorized entry and the advance guidance to the high risk area, It is effective.

FIG. 1 shows the main components of a hazard map-based accident response and accident prevention system according to the present invention.
FIG. 2A illustrates an example of a worker risk case generated when a plurality of tasks are simultaneously performed, in order to explain the function of the incident response module according to the present invention.
FIG. 2B illustrates an analysis tree combining spatial recognition and behavior recognition to explain the function of the incident response module according to the present invention.
FIG. 3 shows a process of constructing an accident prediction prediction model according to the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows the main components of a hazard map-based accident response and accident prevention system according to the present invention.

Referring to FIG. 1, a risk map-based accident response and accident prevention system 100 according to the present invention includes an accident response module 110, an accident prevention module 120, a data warehouse 130, a database unit 140, And a risk map linkage driving module 150. [

The incident response module 110 calculates the risk of the various gas industry facilities and the risk of the worker and combines them into a specific cell of the hazardous facility class map to inform the representative risk according to the grade.

In addition, the incident response module 110 is a real-time incident response technology combined with worker safety. By building a risk map, it is possible to combine the zone-based risk management technique and the state of the worker, And provides a technology that can respond quickly when it occurs.

Here, the risk map is constructed through the calculation of the unit area risk based on the risk matrix, and the unit regional risk for the gas industry facility is determined by the CA (Consequence Analysis) In the case of the present invention, not only the static risk but also the dynamic risk are combined together.

Hereinafter, the meaning of the dynamic risk considering the state of the operator and the process of calculating the dynamic risk will be described in detail.

Since most accidents occur frequently when an operator is working, the risk level is reflected in the area when the worker is administered to the site of the facility and the final risk is indicated.

FIG. 2A illustrates an example of a worker risk case generated when a plurality of tasks are simultaneously performed, in order to explain the function of the incident response module according to the present invention.

Referring to FIG. 2A, according to the present invention, risk per unit is generated according to weather information, facility status information, past accident information, operator's administration and proficiency information based on regional level risk, Each time it enters, a new risk is calculated and a system is shown to allow the operator to return the required response information according to the risk.

In this case, in order to change the risk according to the movement of the worker, the history of the work accident, the work pattern of the worker and the level of the worker are given, and the accident history is stored for each work by grasping the dangerous work and the work procedure, , The worker 's work pattern is derived from both behavior analysis based user analysis and spatial recognition based field analysis.

FIG. 2B illustrates an analysis tree combining spatial recognition and behavior recognition to explain the function of the incident response module according to the present invention.

Referring to FIG. 2B, the analysis tree 30 calculates a corresponding risk using a knowledge tree in which the spatial cognition 10 and the behavior cognition 20 are combined.

In addition, the spatial perception (10) based on-site characteristics analysis is intended to consider the work environment of the worker's work arrangement (13) considering the facility arrangement (11) and the object arrangement (12) Behavioral Awareness (20) based user needs analysis aims at analyzing the relationship between behavioral characteristics (21), proficiency (23) and cause of accident (22) of risk workers.

Spatial cognition (10) and behavior cognition (20), on the other hand, calculate the associated probability through an association process.

The combination of static and dynamic risk is applied as an equal condition to the results of the personal risk mortality rate and the operator's analysis tree, which are the basic risk map index, and the corresponding risk (SR _i ) Is expressed by Equation (1).

Referring to Equation (1), the corresponding risk (SR _i ) is calculated by normalizing the lethal rate f _i and the analysis result k _i in the i-region to 1 and 0 through the total mortality rate and the maximum and minimum values (min) ), And the calculated risk.

Referring again to FIG. 1, the accident prevention module 120 can predict the future risk by applying an intelligent algorithm matching the gas facility domain in which the target facility exists.

Hereinafter, the concept and kind of the accident prediction prediction model applied to the accident prevention module 120 and the method for calculating the prevention risk using the same will be described in detail.

The accident predicting model defines the degree of accident preconditioning as the linkage of the influence factor variables. For example, the influence factors related to the types of risk factors and facility location for the accident precursor are defined as variables, In case the defined influence factors can be observed and measured using IOT (Internet of Things) technology.

The types of accident prediction models are classified into a deterministic model, an artificial intelligence model, and a stochastic model, and the advantages and disadvantages of the respective models are summarized in Table 1 below.

Referring to Table 1, deterministic models use a statistical calculation (such as mean, standard deviation, etc.) to express the relationship between dependent variables and factors that affect the accident history .

The deterministic model can be largely divided into the univariate regression method and the multivariate regression method. The difference between the two methods is that the factor that affects the dependent variable in the deterministic model is one factor or many factors The difference is that, since most predictive dependent variables are associated with more than one influencing factor, multivariate regression analysis is commonly used.

Artificial intelligence models are computationally polynomial models that use computer techniques to take into account variables that affect the prediction of accidents.

In addition, the artificial intelligence technology consists of expert systems, artificial neural networks (ANN), and case based reasoning (CBR). For example, Bayesian Network , Fault tree, Binary Recursive Partitioning, Case-based reasoning (CBR), Knowledge-based systems (CBR) and Artificial neural network.

Stochastic models are used to divide the accident process of a facility into one or more variables, which are used to identify process uncertainties and randomness.

For example, if the instrumentation data is in a rating data format, the discrete-time, risk-based Markovian model is defined as being modeled as a probability of transitioning from one state to another in discrete time, In the case of power generation facilities, it depends on a series of influencing factors such as pressure, temperature and humidity.

The stochastic model includes, for example, a Markovian model, a failure rate model, an ordered probit model, a binary probit model, a Bayesian approach, a semi-Markov model, and a continuous stochastic process model (Gamma process model, Gaussian process model) .

In this case, the representative Markovian model is widely used to model the degradation rate of the facility, for example, a package prediction model, an excellent pipe model, a bridge model and a member model.

On the other hand, as long as gas-related facilities are developed, equipment performance and real-time monitoring and preventive maintenance technologies are developed, it is necessary to change the preventive maintenance standards for gas-related facilities through anticipation of accidents. Be careful.

FIG. 3 shows a process of constructing an accident prediction prediction model according to the present invention.

Referring to FIG. 3, there is a first step (S10) of reviewing and determining prediction data according to the risk factors and the influence factor characteristics.

The risk factors are, for example, explosion and fire. The influencing factors are, for example, failure location, performance degradation mechanism, performance degradation, failure time, detection method, Corrosion and fatigue degradation models.

Depending on the nature of these risk factors and influencing factors, predictive data are determined by examining possible failure rates, wear degradation, corrosion degradation, and fatigue degradation that can be predicted.

Next, there is a second step (S20) of defining influence factors related to the prediction data and selecting final influence factors through correlation analysis.

In more detail, the prediction target data determined in the first step S10 may be, for example, a failure rate, wear, corrosion, fatigue, etc. In this case, influence factors related to respective prediction target data are defined.

For example, the influence factor related to the failure rate can be defined as the number of times of the total inspection, the number of times of failure with respect to the total running time, or the failure time, and the influence factors related to the abrasion are defined by the material properties of the material, the friction coefficient, The influence factors related to corrosion can be defined by the characteristics of material, corrosion rate, humidity, etc. The influence factors related to fatigue can be defined by the fatigue characteristics of the material, the number of repetition of loads, and structural characteristics.

On the other hand, whether the influence factors are actually trending for each predicted data is confirmed through correlation analysis, and the final influence factor is selected.

Next, there is a third step (S30) of constructing a wear, corrosion and fatigue degradation model of a vulnerable position through a deterministic method and a stochastic method, and calculating a failure rate.

As shown in Table 1, there are deterministic models, stochastic models, and artificial intelligence models as predictive models for accident predicting and future states. These deterministic models, stochastic models, and artificial intelligence models are compared and examined Although each model has advantages and disadvantages, it is considered that the most suitable model for the accident predicting model should be considered together with deterministic model and stochastic model (probabilistic model).

In other words, the deterministic method uses a statistical calculation (eg, mean, standard deviation, etc.) to express the relationship between the factors affecting the accident history and the dependent variable, and the stochastic method is a stochastic analysis A method of expressing the predictive model is used in consideration of the uncertainty of factors affecting the accident history.

Hereinafter, the process of calculating wear rate, corrosion and fatigue degradation model and calculation of failure rate of a vulnerable position through deterministic and stochastic methods will be described.

First of all, in case of abrasion, multivariate regression analysis of the deterministic method or a method using the probability distribution in the stochastic method considering the influence factors selected in the second step (S20) The wear degradation model can be constructed.

In the case of corrosion, the characteristics of material, corrosion rate, and humidity are considered as influential factors in the target location, and the risk due to reduction of section can be predicted by applying the proposed degradation models.

In the case of fatigue, it is possible to construct a fatigue degradation model over time through fatigue analysis and multivariate regression analysis of deterministic methods for vulnerable locations, with deterioration resulting from cracking or fracture due to cyclic loading. It is natural that a degradation model can be constructed through the methods described above.

In the case of failure rate, failure rate is calculated by statistical method and probabilistic analysis when there is actual failure data. However, when there is no actual failure data, for example, failure mode / mechanism distribution (1991), NPRD Reliability data such as non-electronic parts reliability data (1995), NCWC-10 (Handbook of Reliability Prediction Procedures for Mechanicla Equipment) and OREDA-2015 (Offshore and Onshore Reliability Data) The rate can be calculated.

Finally, there is a fourth step (S40) of providing an accident predicting model capable of predicting risk in consideration of the deterioration model related to the risk and the accident, and the failure rate.

In this case, when the accident predicting characteristic at the vulnerable position is related to one specific deterioration model, the corresponding model is applied from the deterioration models and the failure rate models constructed in the third step S30 to provide the accident predicting model , And if the degradation model and the failure rate are correlated with each other, they are considered together to provide an accident precursor model.

For example, if the cause of the failure mechanism is related to wear, the most important variable is the initial time at which the failure occurs, or the minimum lifetime.

The minimum lifetime is the minimum time required for the occurrence of the failure, that is, the time required when the wear degradation reaches the limit value. After the minimum lifetime, the failure rate model provides the accident history model constructed with the failure rate over time .

The following Equation (2) shows the precautionary risk (FR _i ) calculated using the accident predicting model.

Here, the prevention of risk (FR _i) is over the maximum value (max) and the minimum value (min) of the total result for the failure rate o _i a result the other of (Advanced Self-Oragnized Map) ASOM results in i area m _i and field techniques, each one And normalization with zero.

Hereinafter, an ASOM (Advanced Self-Organized Map) applied to the present invention is described as an intelligent algorithm for classifying the state of a gas facility.

The Advanced Self-Organized Map (ASOM) is an algorithm that improves the basic SOM (Self-Organized Map). It collects the information received from the target facility-specific sensors, converts the data into data that can be learned by the learning device, The ASOM learning is performed by converting the data into the SOM feature vector. The SOM map is classified by matching the sensor data to the learned ASOM, and the new data is classified using the ASOM map.

Here, the learning of ASOM is used as an input for normalizing the data into a data range for recognizing normal, attention, predicting, and warning patterns and performing the following processing.

First, it has a first step of initializing the connection strength between N input and M output neurons to an arbitrary number of small values.

Next, a second step of presenting the data received from the target facility-specific sensor converted into the SOM characteristic vector as input data.

Next, there is a third step of calculating the degree of similarity using the input data and data of all output layers and the Euclidean distance.

Next, there is a fourth step of outputting data most similar to the input data in the output layer data.

Finally, there is a fifth step of continuously learning and inputting the sensor data.

On the other hand, the main setting conditions for recognizing patterns in ASOM learning process are Window Size and Window Shift Size.

Window size is a function to find a pattern from data measured over time, and it is a technique to construct input data for a few minutes in one time data from one time data rather than one time, as one learning data.

The Window Shift Size is the same as the Window Size in order to find the pattern in the data measured over time, but it differs in that the window jump size is set when the data is constructed in time series.

For example, if the window size is 5 and the window size is 3, data is input every 2 minutes, and 10 minutes of data are combined to form one learning data. When the second learning data is composed, input is made 6 minutes later The data for 10 minutes is summed up into one learning data.

In the learning process of the ASOM, a BMU (Best Matching Unit) having the smallest phase difference of input data should be found. That is, when searching for the node closest to the input data, the Euclide distance method and the Mahalanobis distance method are selected and utilized in the present invention.

Also, to select the neighborhood radius of the BMU, the node with the highest degree of similarity is selected as the neighbor with the neighborhood method and the Mahalanobis distance method.

At the end of the SOM learning process, a cluster is created between nodes with similarities. When the cluster is created, the center of the cluster must be selected and the Predictable Hit Count method of selecting the center of the cluster using the selected number of maps is used.

Referring again to FIG. 1, the data warehouse 130 allows each of the databases 141 to 144 constituting the database unit 140 to be directly used in the incident response module 110 and the accident prevention module 120 Provides a reconstructed information structure.

For example, in order to view the worker and the incident history for the facility in the facility and the data base, two or three or more databases are usually combined, but the safety using the data warehouse 130 according to the present invention Since the management information model is already combined as information, a more efficient search service can be provided.

In other words, in the safety management information model using the data warehouse 130, since the valuable information is generated by the organic combination of the necessary parts of the source database, two to three or more databases are simply combined, The search speed can be remarkably improved as compared with the case where it is slow.

Hereinafter, the concept of the data warehouse and the characteristics of the data structure will be described.

Data or information related to safety management applied to the present invention may be stored for various purposes such as various kinds of data such as hazardous materials collected by a sensor or a meter, state data of facilities, environmental data, operation data, If these data are periodically accumulated, they can be expected to generate data and information for future large-scale safety and disaster management.

The data generated from these safety management industries should be integrated into consistent information, and not only should not be changed once collected and recorded, but also be able to provide the requested information even when the collected data varies with time. Large information structures such as data warehouses must be applied to meet the requirements.

Data warehouses efficiently consolidate, coordinate and manage each and every database management system that is distributed and operated within a large organization and provide the basis for an efficient decision-making system.

The configuration of a data warehouse can consist of, for example, management hardware, management software, extraction, transformation, alignment tools, database marketing systems, meta data, end user access and utilization tools,

Data warehouses are characterized by a collection of subjectoriented, integrated, timevarient, and nonvolatile data to support managerial decision making.

Here, subjectoriented means that the data should be organized by subject, so that end users and analysts who are weak in computation should keep them in an easy to understand format.

Integrated means that all data found in a data warehouse should always be integrated without exception.

Timevarient means that the contents of the data should be maintained even if the contents of the data are changed over time.

Nonvolatile means that data once loaded into the data warehouse is not altered, such as inserts or deletions, except in special cases (such as updates by batch processing).

Referring again to FIG. 1, the database unit 140 includes an accident history DB 141, a material DB 142, an operator characteristic DB 143, and a surrounding DB 144.

The accident history DB 141, for example, accumulates data based on the accident identification and damage information associated with the facility identification number, facility type, facility composition, material, and gas (hazardous material information) information of the target facility.

The material DB 142 accumulates data such as the identification number of the material utilized in the workplace, physical properties, damage effects, extent of damage, and the like.

The worker characteristics DB 143 stores, for example, data on the worker identification number, the worker proficiency, the worker class, and the facility to be managed.

The surrounding environment DB 144 establishes surrounding environmental data of various gas industrial facilities. For example, in addition to the test data used in the Bayesian feasibility study for effective intelligent decision making in the domain of the gas safety facility, the temperature, humidity , Data that can distinguish between underground, ground (indoor), ground (outdoor), and measurement location facilities can be accumulated depending on the illuminance, illumination, and infrared rays.

Referring again to FIG. 1, the risk map linkage module 150 may be expressed by reconstructing a new risk by combining the risk of an incident response with the probability of an accident prediction, which is expressed as a final risk DRi Is expressed.

Referring to equation (3 ) , the final risk ( DRi) Is expressed as a percentage of possible risks at a specific time in the future. Specifically, it is expressed as a product of the corresponding risk (SR _i ) of Equation (1) and the preventive risk (FR _i ) of Equation (2).

Here, the final risk ( DRi) means a risk for the future. As can be estimated from Equation (3 ), the final risk DRi can be calculated as a different value depending on the surrounding environment, the worker's movement,

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments. It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit of the invention.

100: Risk map based accident response and accident prevention system
110: Incident response module
120: accident prevention module
130: Data Warehouse
140:
141: Incident history database (DB)
142: Material Database (DB)
143: Worker characteristics database (DB)
144: environment database (DB)

Claims

An incident response module that calculates the static risk of the gas industry facility and the dynamic risk of the operator and provides the corresponding risk that represents the risk of the risk map for the accident response;
An accident prevention module that establishes an accident predicting model in consideration of a risk and an accident related deterioration model and a failure rate and provides a risk of preventing a future risk prediction through the accident predicting model;
A database unit storing accident history information, material information, operator characteristic information, and surrounding environment information;
A data warehouse for collecting a plurality of risk factors from the database unit and providing a reconstructed information structure for use directly in the incident response module and the accident prevention module without any separate data processing operation; And
And a risk map linkage module coupled with the incident response module and the accident prevention module to combine the risk of an accident response with the probability of an accident prediction to provide a final risk,
The above-
The corresponding risk is calculated by using a knowledge tree which is a combination of spatial perception and behavior perception,
The spatial perception may be determined,
The working environment for the work arrangement of the operator considering the facility arrangement and the object arrangement of the gas facility is taken into consideration,
The above-
The relationship between the risk person's behavioral characteristics and proficiency and the cause of the accident is considered,
The final risk,
Wherein the risk management system is represented by a product of the corresponding risk and the preventive risk.

delete

The method of claim 1, wherein the accident prediction prediction model of the accident prevention module comprises:
A first step of reviewing and determining prediction data according to a risk factor and an influence factor characteristic;
A second step of defining an influence factor related to the prediction data and selecting a final influence factor through correlation analysis;
A third step of constructing a wear, corrosion and fatigue degradation model of a vulnerable position through a deterministic method and a stochastic method, and calculating a failure rate;
And a fourth step of providing an accident predicting model capable of predicting the risk in consideration of the failure rate and the degradation model related to the risk and the accident.

6. The method according to claim 5,
A model of degradation of wear, corrosion and fatigue, and a failure rate.

7. The method according to claim 6,
A risk map based accident response and accident prevention system characterized by application of ASOM (Advanced Self-Organized Map) technique.

6. The method according to claim 5,
Wherein the multivariate regression analysis is used as the deterministic method, and the probability distribution is used as the stochastic method.

2. The apparatus according to claim 1,
An accident history DB, a material DB, a worker characteristic DB, and a surrounding DB.

10. The method according to claim 9,
A risk map-based accident response and accident prevention system, wherein the worker identification number, the worker's proficiency level, the worker's grade, or the worker's facility data are constructed.

delete