JP6937284B2 - Man-hour analysis system - Google Patents

Description

The present invention relates to a man-hour analysis system and its data structure applicable to plant construction estimation and construction management based on past actual work data.

In plant construction, estimates are made based on past actual work data and plan forecasts, adding the knowledge of experienced people. In a certain plant, the site is subdivided into areas, the amount of work (the number of construction parts) in each area is calculated, and the man-hours required for the work according to the difficulty of each work (in the present invention, for construction in construction management). It refers to the required number of workers, for example, if three workers work in one day, it will be three man-hours). Estimating man-hours is an important factor in construction management. At this time, the knowledge of an experienced person is required to estimate the work difficulty level.

However, in the construction of plants with significantly different working conditions, such as the construction at the place where the construction is done for the first time, there was no precedent, and the man-hours increased or decreased due to unexpected factors. For example, construction objects such as plant construction are one-item production and are rarely constructed under the same conditions, and when the particle size of data is large, the amount of data that can be compared under the same conditions is extremely small.
Therefore, the data on which the calculation is based is only past actual work data, but it is necessary to establish a new construction plan and man-hour estimation method for sites that have no actual results and experience based on these data.

Here, as a technique for performing various predictions, normalities, and abnormality judgments using a conventional computer system, Patent Document 1 performs predictions or normality / abnormality diagnosis by the minimum square method with data in a database (hereinafter, may be simply referred to as DB). It provides a calculation method.
However, Patent Document 1 only discloses a method for constructing a calculation method, and does not disclose a specific method for utilizing a database or an evaluation method for the obtained output result.

Japanese Patent No. 5510642

In view of the above-mentioned problems of the prior art, the present invention makes it possible to improve and rationalize the accuracy of man-hour estimation even if there is a large uncertain factor that cannot predict the man-hour required for construction, and is applied to various estimation calculations and construction management. The purpose is to provide a possible man-hour analysis system and its data structure.

The present invention is a man-hour analysis system that generates a formula for calculating man-hours required for work based on past actual work data as a first means for solving the above problems.
A performance work database that is connected to the work terminal via an internal network and stores the performance work data including past work daily report data and work records, and a construction management database that stores product information, drawing information, and worker information. ,
An external database that is connected to the work terminal via the internal network and further via an external network and stores at least weather information and economic information.
Connected to the work terminal, the reference man-hours obtained by dividing the man-hours of the work by the physical unit of the work were created, and the internal factors obtained through the internal network of the work and the internal factors obtained via the external network were obtained. A data creation server that creates a man-hour fluctuation factor to which an external factor is added, and the reference man-hour and the man-hour fluctuation factor stored in the data creation server are read out, and the man-hour fluctuation that is considered to contribute to the fluctuation of the reference man-hour. A man-hour calculation server that creates an evaluation formula that estimates man-hour fluctuations by automatically generating and extracting feature quantities that are a combination of multiple factors that have a high correlation with the standard man-hours from among the factors by machine learning.
The purpose is to provide a man-hour analysis system characterized by being equipped with.
According to the first means described above, it is necessary to provide a method for newly estimating the fluctuation factor of the actual man-hours and the degree of influence thereof from the tendency of the past actual work data, which could not be predicted by the conventional method or even an experienced person, and is necessary for the construction. The man-hour estimation accuracy can be improved and rationalized, and it can be applied to various estimation calculations and construction management.

According to the present invention, as a second means for solving the above-mentioned problems, in the first means, the man-hour calculation server clearly correlates with the man-hour fluctuation when the evaluation formula is less than the coefficient of determination. The purpose is to provide a man-hour analysis system characterized by normalizing and removing factors by statistical processing.
According to the second means described above, it becomes possible to clarify the fluctuation of the clear elements of the actual man-hours and the man-hour calculation formula with respect to the data excluded from the actual man-hours in advance. In man-hour prediction, the calculation accuracy can be improved by combining the elements divided in advance and the man-hour calculation formula used in this method.

According to the present invention, as a fourth means for solving the above problems, in any one of the first to third means, the internal factor is the work record of the actual work database and the product accumulated in the construction management database. An object of the present invention is to provide a man-hour analysis system characterized in that it is one or more of information, drawing information, and worker information.
According to the fourth means, it is possible to find a factor that may affect the man-hour fluctuation that could not be predicted by the conventional method or even an experienced person and add it to the man-hour calculation formula.

According to the present invention, as a fifth means for solving the above problems, in any one of the first to fourth means, the external factor is the weather information, economic information, and employment conditions accumulated in the external data. An object of the present invention is to provide a man-hour analysis system characterized by having one or more of them.
According to the fifth means, it is possible to find factors that affect the man-hour fluctuation, which could not be assumed by the conventional method or an experienced person, and add them to the man-hour calculation formula.

According to the present invention, as a sixth means for solving the above problem, in any one of the first to fifth means, the man-hour calculation server uses the reference man-hours stored in the data creation server and the reference man-hours. The man-hour fluctuation factors are read out, and the feature quantity, which is a combination of a plurality of factors having a high correlation with the reference man-hours, is automatically generated / extracted from the man-hour fluctuation factors considered to contribute to the fluctuation of the reference man-hours by machine learning. An object of the present invention is to provide a man-hour analysis system characterized by being a factor extraction type artificial intelligence that creates an evaluation formula expressing the relationship together with the feature amount.
According to the sixth means, by machine learning using AI (artificial intelligence), it is possible to automatically extract which combination of man-hour fluctuation factors matches the explanation of man-hour fluctuation, and it can be assumed by the conventional method or an experienced person. It is possible to find out the factors that affect the man-hour fluctuations that could not be done, and add them to the man-hour calculation formula to create a highly accurate estimate with less man-hour fluctuations.

The present invention is a man-hour analysis method using a man-hour analysis system including a database, a data creation server, and a man-hour calculation server as a sixth means for solving the above problems.
The database is one of the actual man-hours including the past daily work report data and work records, which are the results of plant construction, product information, drawing information, worker information, weather information, and economic information. Accumulate more than one
The data creation server encloses the work numbers to be managed from the accumulated information in the database, and the man-hour data for each element work to be predicted, the element work included in the work number, and the work for each element work. Create factor data showing the conditions and
In the man-hour calculation server, an evaluation formula for estimating man-hour fluctuation is created by extracting the factor data having a large influence on the man-hour data by machine learning by AI, and the value calculated by the evaluation formula and the actual man-hour An object of the present invention is to provide a man-hour analysis method for calculating man-hours based on the evaluation formula when the relationship is equal to or higher than a predetermined determination coefficient.
According to the above-mentioned seventh means, a method for newly estimating the fluctuation factor of the actual man-hours and the degree of influence thereof, which could not be predicted by the conventional method or even an experienced person, from the tendency of the past actual work data is provided, which is necessary for the construction. The man-hour estimation accuracy can be improved and rationalized, and it can be applied to various estimation calculations and construction management.
The present invention is a man-hour analysis method using a man-hour analysis system including a database, a data creation server, and a man-hour calculation server as a seventh means for solving the above problems.
The database is one of the actual man-hours including the past daily work report data and work records, which are the results of plant construction, product information, drawing information, worker information, weather information, and economic information. Accumulate more than one
The data creation server encloses the work numbers to be managed from the accumulated information in the database, and the man-hour data for each element work to be predicted, the element work included in the work number, and the work for each element work. Create factor data showing the conditions and
In the man-hour calculation server, an evaluation formula for estimating man-hour fluctuation is created by extracting the factor data having a large influence on the man-hour data by machine learning by AI, and the value calculated by the evaluation formula and the actual man-hour An object of the present invention is to provide a man-hour analysis method for calculating man-hours based on a reference man-hour calculation formula represented by a product of the evaluation formula and a linear regression formula when the relationship is less than a predetermined determination coefficient.

According to the present invention based on the above configuration, a method for newly estimating the fluctuation factor of the actual man-hours and the degree of influence thereof, which could not be predicted by the conventional method or even an experienced person, from the tendency of the past actual data is provided, and the man-hours required for construction are provided. The estimation accuracy can be improved and rationalized, and it can be applied to various estimation calculations and construction management. In addition, the man-hour calculation server using AI can extract potential factors and create a highly accurate estimate with factors that could not be predicted by conventional estimates.

It is a block diagram of the man-hour analysis system of this invention. It is a processing flow diagram of the man-hour analysis system of this invention. It is explanatory drawing of the standard man-hours. It is explanatory drawing of the man-hour fluctuation factor. It is explanatory drawing of the data structure of man-hour analysis. It is explanatory drawing of the man-hour fluctuation factor which is clear when the pipe groove alignment work is taken as an example. It is explanatory drawing which compared the actual standard man-hours (actual data) and the standard man-hours (man-hour calculation data) calculated from the man-hour calculation formula by the conventional statistics at the site where the man-hours are totaled for each work. It is a table which shows the relationship between the value of a coefficient of determination and the strength of correlation. It is explanatory drawing of the man-hour calculation formula by AT / H for the said pipe groove alignment work. It is explanatory drawing which compared the actual standard man-hour and the standard man-hour calculated from the man-hour calculation formula by the conventional statistics and the man-hour calculation formula by a man-hour calculation server. It is explanatory drawing of man-hour data and factor data. It is explanatory drawing of the man-hour data and factor data normalized with the average diameter as a clear influence factor.

An embodiment of the man-hour analysis system and its data structure of the present invention will be described in detail below with reference to the accompanying drawings.

[Man-hour analysis system 10]
FIG. 1 is a block diagram of the man-hour analysis system of the present invention. As shown in the figure, the man-hour analysis system 10 of the present invention mainly includes a performance work database 20, a construction management database 30, an external database 40, a data creation server 50, and a man-hour calculation server 60.

In the effort analysis system 10 of the present embodiment, the data creation server 50, the effort calculation server 60, and the input / output interface 70 are connected to the user's work terminal 12, and the actual work database 20 and construction management are further performed via the internal network 14. It is connected to the database 30 and further connected to the external database 40 via the external network 16. The user's work terminal 12 has a well-known configuration such as a CPU, a memory, a keyboard, a touch screen, a microphone, and a speaker.
The actual work database 20 is a database in which actual man-hours including data and work records of past daily work reports, which are the actual results of plant construction, are accumulated.

The construction management database 30 is a database that stores product information, drawing information, and worker information.
Product information is information such as diameter, length, material, weight, etc. in the case of products (components) required for plant construction, for example, piping.
The drawing information is information such as a work place of plant construction, an area of a work area, and a shape.
Worker information is information such as the career (years of experience) and qualifications of workers involved in plant construction. The data from the above-mentioned performance work database and construction management database are internal factors.

The external database 40 is a database that stores weather information, economic information, and the like. Of these, economic information and the like are various external factors including, for example, exchange rates, international affairs, economic conditions and employment conditions.
The data creation server 50 is a server that converts the format of data used in the man-hour calculation formula for obtaining the man-hours required for plant construction.
Here, the man-hour calculation formula is a formula for calculating the man-hours per unit physical quantity for each work when the man-hours are managed for each work. This man-hour calculation formula can be created by general calculation program software such as Microsoft Excel or open source R. In this embodiment, it is created by the man-hour calculation server described later.

FIG. 3 is an explanatory diagram of the reference man-hours. The standard man-hour is, for example, in the case of pipe groove alignment work (for example, the work of lifting from carry-in to a predetermined position, aligning the groove and welding, the same applies hereinafter), carrying in, lifting, aligning the groove, Divided into welding (here, element work), and in the work performed for each element work (work number, work name), the man-hours required for the work are divided into the physical unit of the work (parts if carried in). It is a value divided by the number of pipe connection points (points) for each weight and groove alignment. The larger the value, the more man-hours are required for the work per unit, and conversely, the smaller the value, the more man-hours are required for the work. It shows that the man-hours were small. This standard man-hour (man-hour / point) can be obtained from the actual work data. By using the standard man-hours, it is possible to create a man-hour calculation formula by separating the man-hours that fluctuate due to various factors and the man-hours for which the quantity is clear in advance.

The data creation server 50 of the present embodiment performs data format conversion using the actual work database 20 to create a reference man-hour.
Then, the construction management database 30 and the external database 40 are used to perform data format conversion to create a man-hour fluctuation factor.
FIG. 4 is an explanatory diagram of man-hour fluctuation factors.
The man-hour fluctuation factor is factor data that increases or decreases the man-hour. Specifically, it is data showing a factor item and its definition (range). For example, in the case of an object size of a factor item, the definition is a factor that defines a size such as a pipe diameter and a device weight. The work position information is defined by the height of the work place from the floor and the like. The work space density is a value that defines the work space density (work object capacity / space volume, etc.). The work area information is a value that defines the complexity of the area shape (area variable, area area, area circumference / area, etc.). Worker information is information that defines the conditions (affiliation, qualification, etc.) of workers engaged in work. The meteorological conditions are the meteorological conditions at the time of work implementation. Economic information is various factors such as fluctuations in employment conditions caused by economic conditions. Depending on the unemployment rate, etc., the skill level of working in the field may differ.

[Data structure of man-hour analysis]
FIG. 5 is an explanatory diagram of a data structure for man-hour analysis. The data structure used for man-hour analysis is roughly divided into two configurations, one is the reference man-hour data for each work to be predicted shown in FIG. 3 (hereinafter referred to as man-hour data in the present invention), and the other is. It is the condition data of the work object shown in FIG. 4 (hereinafter, referred to as factor data in the present invention).
The man-hour data has a structure in which the unit to be predicted (in this embodiment, by work) and the data of the value to be predicted are put together.

The factor data has a structure in which various factor data that are conditions for man-hour fluctuation are put together, and the number of items (number of factors) is enormous compared to the man-hour data. As an example (see FIG. 4), the man-hour data for the pipe groove alignment work is A1: pipe No., A2: diameter, A3: thickness as data (specific example) for each connection point for the object size (No1). (Thickness), A4: Length of pipes constituting the connection point, A5: Material, A6: Weight, A7: Shape of straight pipe, curved pipe, etc., A8: Includes data such as groove type of connection point ..
Regarding work position information (No. 2), B1: the floor of the building through which the connecting pipes pass, B2: the height of the connection point from the floor, B3: the opening (carrying entrance) for carrying in the pipes, B4: the room through which the pipes pass. Includes any of the necessary information from the opening size of.
It has data of all factors (Nos. 1 to 7 ... In FIG. 4) that can be considered in the same manner as a group. In addition, since these data groups include factors depending on the time period, the data are summarized in chronological order in this example. A certain work area has a structure in which data is stored for each work period. The basic analysis method is to perform diffraction analysis of fluctuations in man-hour data using the conditions of factor data.

Here, conventionally, the calculation program software of the man-hour calculation formula has been performed as follows. FIG. 6 is an explanatory diagram of factors that cause fluctuations in man-hours, which is obvious when the pipe groove alignment work is taken as an example. The graph in the figure shows the standard man-hours related to the pipe groove alignment work, the average diameter (mm) of the pipes within the management target range such as the area, the standard man-hours related to the pipe groove alignment work, and the work period (month from the start of construction). Number), The relationship between the standard man-hours related to the piping groove alignment work and the ratio of the number of workers (%) to the number of workers who have been engaged in the field for one year or more is shown. Y = ax + b (the standard man-hours increase as the diameter increases), Y = ax ² + bx + c (the amount of work is small at the start of construction, and the standard man-hours are relatively high because it includes time for preparation and familiarization, etc. The most efficient, the amount of work will be reduced again at the end of construction and the standard man-hours will be relatively high), Y = -ax + b (the standard man-hours will decrease as the number of workers accustomed to the site increases), etc. It can be considered that there is a certain correlation depending on the content and content.

FIG. 7 is an explanatory diagram comparing the actual standard man-hours (actual data) and the standard man-hours (man-hour calculation data) calculated from the man-hour calculation formula based on the conventional statistics at the site where the man-hours are totaled for each work. The vertical axis of the figure shows the standard man-hour fluctuation ratio, and the horizontal axis shows the work number (the one enclosed by the management target such as the area). The round plot shows the actual data, and the square plot shows the standard calculation data calculated from the man-hour calculation formula based on the conventional statistics. When the man-hour calculation data created by the conventional statistics using only the three obvious man-hour fluctuation factors shown in FIG. 6 and the actual data were compared, many differences still appeared between the two. In this way, conventional statistical predictions alone are the limit of estimation to this extent, and further predictions have to rely on the knowledge of experts.

The work performance analysis system of the present invention uses the man-hour calculation server 60 to create a man-hour calculation formula for obtaining an estimated value of the actual man-hours using the conventional statistical calculation software, and uses it as an index for estimation and construction management.
The man-hour calculation server 60 performs machine learning using AI (artificial intelligence) regarding the correlation and combination of the reference man-hours which are man-hour data and the man-hour fluctuation factors which are factor data.
This AI technology uses Hitachi AI Technology / H (hereinafter, simply referred to as AT / H), which is an artificial intelligence originally developed by Hitachi. AT / H is an object from a group of explanatory variables (man-hour fluctuation factors of the present embodiment) that are considered to contribute to fluctuations of a certain objective variable (man-hours of the present embodiment) when applied to estimate the degree of influence of man-hour fluctuation factors. This is a factor-extracting artificial intelligence that automatically generates and extracts combinations (features) of multiple factors that are highly correlated with variables by machine learning, and formulates the relationships.

AT / H uses the aggregated data of the actual man-hours for each area in the construction and the condition data for each object of the construction objects in the area, which is the breakdown of the actual man-hours, to determine the feature amount that fluctuates the actual man-hours. It is automatically created and selected from various variable factors. At this time, a formula that estimates the man-hour fluctuation by grouping the features of similar factor groups that have a high correlation with the fluctuation of the actual man-hours, and selecting the appropriate feature from the group that has a high influence on the man-hour fluctuation. To derive. In addition, the degree of influence can be quantitatively estimated from the coefficients of the derived equation.
Here, the factors estimated by AT / H are the features of the higher ranks with high influence (for example, three, which can be changed by setting). Therefore, when all the factors are calculated only by AT / H, a person experiences it. Those that can be relatively easily determined from the above are preferentially extracted as factors with high correlation. On the other hand, it is presumed that potential factors that are difficult for humans to judge are less correlated than others and are not reflected in the evaluation formula. Therefore, the man-hour fluctuation and the obvious factor are analyzed in advance by conventional statistical processing such as linear regression, and the standard man-hour is divided by the regression equation and normalized as the new standard man-hour. By letting AT / H process using this, the influence of obvious factors can be excluded from the scope of analysis.

New standard man-hours (man-hours / points) = standard man-hours (man-hours / points) / (α x factor 1 x β) ... (1)
As an example, in the case of pipe groove alignment work, the larger the pipe diameter, the larger the man-hours required for the work. By using the value obtained by normalizing the man-hours for the pipe diameter as the reference man-hour for analysis, it becomes possible to estimate the man-hour fluctuation factors other than the pipe diameter by AT / H.
The man-hour calculation formula by AT / H using the new standard man-hour can be expressed by the following formula.

New standard man-hours (man-hours / points) = (a1 x feature 1 + a2 x feature 2 + a3 x feature 3 + b) ... (2)
From the above (1) and (2), the prediction formula for the standard man-hours is as follows.
Standard man-hour prediction formula (man-hours / points) = (α x factor 1 x β) x (a1 x feature 1 + a2 x feature 2 + a3 x feature 3 + b) ... (3)
Here, factor 1 is obtained by a regression equation based on an obvious factor (pipe diameter), and the feature amount is obtained by an evaluation formula based on AI.

The above equation (3) is an example when only the pipe diameter is the normalization target and the feature amount by AT / H is three. Depends on the number of data to be analyzed, but can generally be expressed by the following equation.
Reference man-hour prediction formula (man-hours / points) = (regression formula by factor 1) × (regression formula by factor 2) × ・ ・ × (regression formula by factor n) × (a1 × feature amount 1 + a2 × feature amount 2 + a3 × feature amount 3 + ・ ・ + an × feature amount n + b) ・ ・ ・ (4)
If there is a lot of data on actual man-hours and various information can be obtained from the construction management DB30 and the external DB40, the unemployment rate at the construction site will be affected by the above formula (4), as well as the weather conditions and, in some cases, the economic situation. (If the unemployment rate at the construction site is high, there is a possibility that there are surplus skilled craftsmen, and it can be predicted that there will be many skilled workers and the man-hours will be reduced.) Impact on man-hour fluctuations, etc. May be derived.

As a criterion for judging the quality of the obtained man-hour calculation formula, the coefficient of determination between the man-hours calculated in the man-hour calculation and the actual man-hours was evaluated. The coefficient of determination is the square of the correlation coefficient between the result of the calculation formula and the actual value. FIG. 8 is a table showing the relationship between the value of the coefficient of determination and the strength of correlation. When the coefficient of determination is 16% or more, it is generally said that there is a relatively correlation between the actual result and the calculation formula.
FIG. 9 is an explanatory diagram of a man-hour calculation formula by AT / H for the pipe groove alignment work.
In this figure, as factors that have a clear influence on man-hour fluctuations in advance (see Fig. 6), the degree of influence is statistically normalized for the construction time and the worker's on-site engagement period in addition to the pipe diameter, and the degree of influence is sequentially normalized. The calculation is shown excluding. The standard man-hour calculation formulas that have been normalized by the pipe diameter, timing, and worker engagement period are sequentially calculated by regression formulas based on conventional statistics, and the formulas here are omitted.

The top feature quantities 1-3, which are extracted by AT / H and have a high influence on the man-hour variation, are calculated by the linear equation of the coefficient and the intercept. The height of the work place, the shape of the area, the area of the area, and the degree of density are all quantitative information derived by defining the calculation conditions.
From feature 1, the coefficient is 2.88 and the intercept is 0.85 when the work place height is relatively high and the shape of the work area is complicated, so the standard man-hours are 3.7 times as many. It can be judged that there was a tendency to become, and the work corresponding to these was 10 work in the work to be analyzed. Similarly, when the area area is relatively small and the density is high as the feature amount 2, there is a tendency for the man-hours to increase 4.0 times, and the corresponding work is 1. From the feature amount 3, the man-hours tended to increase by 1.5 times in the place where the area area was medium, and the corresponding work was 15 works. It was found that the features of each work that did not correspond to the features 1-3 became 0, and the man-hours tended to decrease 0.85 times from the intercept. The numerical value of the feature amount is 1 if all the objects meet the conditions, 0 if all the objects are not met, and 0.5 if half of the features are met.

In this example, in addition to the influential factors (variable factors) considered as shown in Fig. 6 as the influence on the actual man-hours is clear, the combination of man-hour fluctuation factors from various influential factors of the factor data as shown in Fig. 4 By automatically creating features (combinations of factors) that explain enormous man-hour fluctuations, and grouping combinations that have similar effects as similar features, there is no statistical contradiction (multicollinearity). (Avoid) The combination of man-hour fluctuation factors is automatically extracted. As a result, the man-hour calculation server 60 can specifically predict the man-hours by the standard man-hour calculation formula based on the conventional statistics × the man-hour calculation formula using AI.
As a result of automatically extracting the man-hour fluctuation factors that affect the work man-hours by the above-mentioned man-hour calculation server 60, it can be understood that the reference man-hours tend to increase due to the man-hour fluctuation factors of the work position information and the work space density, for example. become.

FIG. 10 is an explanatory diagram comparing the actual standard man-hours with the standard man-hours calculated from the man-hour calculation formula based on the conventional statistics and the man-hour calculation formula using the man-hour calculation server. The vertical axis of the figure shows the standard man-hour fluctuation ratio, and the horizontal axis shows the work number. The round plot shows the actual standard man-hours (actual data), and the square plot shows the standard man-hours calculated from the man-hour calculation formula by the man-hour calculation server of the present embodiment. AT / H automatically extracts factors of man-hour fluctuations other than pipe diameter, construction time, and worker's on-site working hours, and combines them with evaluation by conventional statistical processing. As a result, the difference between the man-hour calculation formula based on the conventional statistics and the man-hour calculation formula based on AI was smaller than the man-hour calculation formula based on the conventional statistics shown in FIG. 7.
Further, the man-hour calculation server 60 defines in advance a threshold value of the actual standard man-hours and the coefficient of determination of the man-hour fluctuation formula (actual standard man-hour / value of the man-hour fluctuation formula × 100) (%). The larger the value of this coefficient of determination, the smaller the correlation with the standard man-hours, and the smaller the value, the smaller the correlation with the standard man-hours. In the present embodiment, the threshold value of the coefficient of determination is set to 40% as an example. If it is 40% or more, the man-hour calculation formula is determined, and if it is less than 40%, an explanatory formula using influential factors is created. Then, a new automatic extraction is performed again by the man-hour calculation server 60. It should be noted that this threshold value can be set higher for work in which the variation in man-hours is small.

[Man-hour analysis method]
The man-hour analysis method of the present invention based on the above configuration will be described below.
FIG. 2 is a processing flow diagram of the man-hour analysis system of the present invention.
<Create data for AI calculation>
FIG. 11 is an explanatory diagram of man-hour data and factor data.
The data structure used for the man-hour analysis consists of two configurations, one of which is related to the man-hour data for each element work to be predicted, which is enclosed by the work number to be managed. In addition to the man-hour data, there is a groove alignment point for piping groove alignment work as the work volume for each work number so that comparisons can be made for each work number, and per unit quantity obtained by dividing the man-hours by this quantity. Data of standard man-hours (man-hours / points) representing the man-hours of The other is factor data showing all the elemental works included in the work number and the work conditions for each elemental work.

Here, the factor data has a structure that summarizes the conditions of the factors that are considered to fluctuate the man-hours of each element work, and has a huge number of items (element work number and man-hour fluctuation factor) as compared with the man-hour data. These data are acquired from the construction management DB 30 and the external DB 40 of FIG. From the construction management DB 30, conditions of all factors that can be considered from the size and shape of the object, the material, the connection point information, the information on the work position, the qualification information of the worker, and the like may be registered.
Furthermore, from the external DB40, the unemployment rate, the acceptance rate of foreign workers, etc., which may affect the work at the relevant construction site, depending on the weather information at the time of the implementation of the target elemental work and the economic situation at the time of the work implementation. Factor information may be registered.
If information on all factors that can be considered and data can be acquired is registered, AT / H will automatically generate and extract features that are a combination of factors that match the explanation of man-hour fluctuations according to the number of data. ..

Step 1: The data creation server 50 performs data format conversion (prediction target data) based on the data in the actual work database 20. Specifically, the actual man-hours that are the man-hour data used in the process calculation formula and the actual standard man-hours expressed by the actual man-hours per unit weight are calculated.
Step 2: Next, the data creation server 50 performs data format conversion (prediction target condition data) based on the data of the construction management database 30 and the external database 40. Specifically, a list of actual conditions of man-hour fluctuation factors used in the process calculation formula of the structure that becomes factor data is created.

<Extraction of factors by AI and creation of man-hour calculation formula>
Step 3: In the man-hour calculation server 60, a huge amount of feature data that combines a plurality of man-hour fluctuation factors is automatically generated by machine learning by AI (artificial intelligence), and which feature combination is selected from those feature amounts. Automatically extract whether it matches the explanation of man-hour fluctuation.
Step 4: Then, a man-hour calculation formula based on the extracted man-hour fluctuation factors is created (man-hour calculation formula based on conventional statistics × man-hour calculation formula based on AI). If the influencing factors with a clearly high correlation as shown in FIG. 7 have not been selected, AI may automatically extract the factors having a strong correlation at this stage and extract only them.
Step 5: Next, the coefficient of determination of the value of the performance reference process and the created man-hour calculation formula is calculated, and it is determined whether or not the value is equal to or greater than the threshold value.
Step 6: If it is equal to or more than a predetermined threshold value, it is determined as a man-hour calculation formula close to the actual standard man-hours.

Step 7: On the other hand, if the threshold value is not reached, a factor having a strong influence extracted by AI is selected. In the case of the pipe groove alignment work, in addition to the pipe diameter, the construction time, the worker's on-site engagement period, etc.
Since the factors estimated by AT / H estimate the high-ranking features with high influence, when all the factors are calculated only by AT / H, those that can be relatively easily judged by humans from experience have high correlation. It is preferentially extracted as a factor. On the other hand, potential factors that are difficult for humans to judge may be less correlated than others and may not be reflected in the evaluation formula. Therefore, in such a case, the factors whose correlation with the man-hour fluctuation is clear are out of the range by analyzing the degree of influence by conventional statistical processing such as linear regression.

<Creation of man-hour calculation formula by conventional statistical processing>
Step 8: Determine if the impact is a clarified factor. It may be a factor having a high correlation with man-hour fluctuations that can be relatively easily determined by humans from experience, or it may be a factor first derived by AT / H. In the case of the pipe groove alignment work, in addition to the pipe diameter, the construction time and the worker's on-site engagement period are used. If there is no factor, the process proceeds to step 1.
Step 9: If there is a factor in step 1 (Yes), the degree of influence is quantified by conventional statistical processing.
Step 10: Create new standard man-hour data by removing the influence of the revealed factors (data normalization). After that, the process proceeds to step 1.

FIG. 12 is an explanatory diagram of man-hour data and factor data normalized with the average diameter as a clear influencing factor. In the man-hour data shown in the figure, a new standard man-hour is displayed in which the standard man-hour before removing the influential factor is normalized and updated.
According to the present invention as described above, the man-hour calculation formula derived by AT / H does not predict the man-hours of unknown work, and selects the one having a high contribution to the man-hour fluctuation from the man-hour fluctuation factors taken up in advance. If data on other factors can be collected, the coefficient of determination may increase. It is effective that AI technology automatically selects the work that humans had to manually confirm each factor in the conventional statistical processing.

The preferred embodiment of the present invention has been described above. However, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the gist of the present invention.
Further, the present invention is not limited to the combinations shown in the embodiments, and can be implemented by various combinations.

The man-hour analysis system of the present invention has industrial applicability, especially in estimating plant construction.

10 Man-hour analysis system 12 Work terminal 14 Internal network 16 External network 20 Actual work database 30 Construction management database 40 External database 50 Data creation server 60 Man-hour calculation server 70 I / O interface

Claims (7)

In a man-hour analysis system that generates a formula for calculating the man-hours required for work based on past actual work data
A performance work database that is connected to the work terminal via an internal network and stores the performance work data including past work daily report data and work records, and a construction management database that stores product information, drawing information, and worker information. ,
An external database that is connected to the work terminal via the internal network and further via an external network and stores at least weather information and economic information.
Connected to the work terminal, the reference man-hours obtained by dividing the man-hours of the work by the physical unit of the work were created, and the internal factors obtained through the internal network of the work and the internal factors obtained via the external network were obtained. A data creation server that creates a man-hour fluctuation factor to which an external factor is added, and the reference man-hour and the man-hour fluctuation factor stored in the data creation server are read out, and the man-hour fluctuation that is considered to contribute to the fluctuation of the reference man-hour. A man-hour calculation server that creates an evaluation formula that estimates man-hour fluctuations by automatically generating and extracting feature quantities that are a combination of multiple factors that have a high correlation with the standard man-hours from among the factors by machine learning.
Man-hour analysis system characterized by being equipped with. In the man-hour analysis system according to claim 1,
The man-hour calculation server extracts by AI based on the factor data after normalizing and removing the factors whose correlation with the man-hour fluctuation is clear by statistical processing when the evaluation formula does not reach the predetermined coefficient of determination. A man-hour analysis system characterized in that man-hours are calculated based on a calculation formula of a reference man-hour represented by a product of an evaluation formula using a feature amount and a linear regression formula using the above-mentioned obvious factor. In the man-hour analysis system according to claim 1,
The man-hour calculation server calculates man-hours based on the evaluation formula when the value of the evaluation formula for calculating the reference man-hour as the sum of the terms obtained by multiplying the feature amount extracted by AI by the coefficient satisfies a predetermined determination coefficient. Man-hour analysis system characterized by In the man-hour analysis system according to any one of claims 1 to 3,
The man-hour analysis system is characterized in that the internal factor is any one or more of the work record of the record work database, the product information accumulated in the construction management database, the drawing information, and the worker information. In the man-hour analysis system according to any one of claims 1 to 4.
The man-hour analysis system characterized in that the external factor is any one or more of weather information, economic information, and employment conditions accumulated in an external database. It is a man-hour analysis method using a man-hour analysis system consisting of a database, a data creation server, and a man-hour calculation server.
The database is one of the actual man-hours including the past daily work report data and work records, which are the results of plant construction, product information, drawing information, worker information, weather information, and economic information. Accumulate more than one
The data creation server encloses the work numbers to be managed from the accumulated information in the database, and the man-hour data for each element work to be predicted, the element work included in the work number, and the work for each element work. Create factor data showing the conditions and
In the man-hour calculation server, an evaluation formula for estimating man-hour fluctuation is created by extracting the factor data having a large influence on the man-hour data by machine learning by AI, and the value calculated by the evaluation formula and the actual man-hour A man-hour analysis method for calculating man-hours based on the evaluation formula when the relationship is equal to or greater than a predetermined determination coefficient. It is a man-hour analysis method using a man-hour analysis system consisting of a database, a data creation server, and a man-hour calculation server.
The database is one of the actual man-hours including the past daily work report data and work records, which are the results of plant construction, product information, drawing information, worker information, weather information, and economic information. Accumulate more than one
The data creation server encloses the work numbers to be managed from the accumulated information in the database, and the man-hour data for each element work to be predicted, the element work included in the work number, and the work for each element work. Create factor data showing the conditions and
In the man-hour calculation server, an evaluation formula for estimating man-hour fluctuation is created by extracting the factor data having a large influence on the man-hour data by machine learning by AI, and the value calculated by the evaluation formula and the actual man-hour A man-hour analysis method for calculating man-hours based on a standard man-hour calculation formula represented by the product of the evaluation formula and a linear regression formula when the relationship is less than a predetermined determination coefficient.

Images