An Effective Optimization of Time and Cost Estimation for Prefabrication Construction Management Using Artificial Neural Networks

A country's economy is controlled by several aspects, i.e. population size, manufacturing practices, agriculture, government policy, people's culture, schooling, infrastructure, etc. In order to fulfill basic needs including food, shelter and clothes, citizens participate in different activities including agriculture, housing and the textile industry. Each of the industries listed above is assisted by numerous other industries. For example, machine tools, agriculture and chemical industries help a textile industry. A substantial part of the commodity in each sector is used by the public and a proportion is fed to other companies as raw materials or machinery. In addition to these manufacturing practices, educational establishments are involved with and preparing appropriate workers for these sectors. In addition to these operations, there are various service organisations, including hospitals, travel, postal facilities, legal services, insurance agencies, financial firms, etc.

Optimisation, including the optimization or minimisation of a real function, is accomplished by manually extracting the input values from within the permissible range and determining the function value. The widespread optimization theory and techniques are a vast field of applied mathematics among several formulations. In general, optimization seeks the best available values for certain objective functions within a certain range of specified domains or restrictions. Optimization often requires various forms of objective features and different domain forms. The emphasis of this analysis is on optimization strategies to reduce prefabricated buildings time and cost.

Based on existing literature, the present state of the building sector and the usage of prefabricated technologies was analysed. The latest developments in housing design and prefabrication technologies utilized by construction firms are discussed in this report. The appraisal practices for construction management are often examined and reflected on. A favourable construction management approach was then suggested. Prefabrication construction is expected by the design sector and steps are often implemented to boost the adverse condition. The present study shows why Indian construction industry is lagging behind the developed countries in the usage of state-of-the-art information technology and offers guidance for enhancing performance. The selection of the most beneficial construction equipment is important for savings by minimizing costs and simplifying and encouraging construction. There are suitable technical models that can be used for decision-making in the most favourable building process. The use and production of prefabrication technology in civil engineering has been enhanced dramatically in developed countries relative to the last decades, but it cannot yet be seen as adequate [1].

With its excellent production methods for the efficiency limits of the prefabricated building phase ANN would in future create additional amazing advance methodologies for achieving error reductions. The policies and guidance of both the central governments and the state governments promote the inclusion, alignment and regulation of all the nation's activities in order to optimise the nation's progress. It is also evident that any organisation of the country is required to play a competitive position in order to increase its efficiency for their existence.

A scheme is time-bound and employs tremendous resources of individuals, materials and machinery. The currency is millions and trillions. Technical breakthroughs also affected the building industry tremendously. In multidimensional terms, high-rise buildings, industrial plants and infrastructural facilities have been built. The construction industry is a major contributor to a country's growth, both domestic and foreign. This sector produces more workers than any other industries. Due to political and economic pressures on the construction industry, such as privatization of state-owned corporations, the opening of the Indian market, vigorous reconstruction work after catastrophe, intensive road building, etc., Indian construction firms have experienced extreme shift for two decades. These turbulent developments have affected building industries in numerous ways. Many formerly powerful Indian firms have vanished from the sector, and others have turned their management and design processes into prefabricated framework technology. Prefabricated architecture involves producing construction components in a production facility, transporting assembled or semi-finished components to the building site and eventually installing such components in a building [2]. The prefabrication process requires the manufacture and transfer of precast parts to the building site [3].

1.1 Prefabrication need

The industrialization of the building sector means a revolution in the whole design phase. The desire for safer, quicker and commercially feasible construction has contributed to a large market footprint for the pre-casting technology. In order to avoid on-site adjustments, careful preparation, thorough design and scheduling of the project before building activities must be taken into account. In several countries and territories, prefabrication technology has typically been utilised when a certain amount of building materials are made in a regulated setting, transported to building sites and installed in houses. The prefabricated housing units have been improved in such a way that they cannot even be separated from those housed in traditional buildings. In challenging weather environments, prefabricated structures can be necessary to increase the building pace and also to reduce the waste of the construction material by using mass manufacturing under control over the material. This criterion has now been increased to install broad span beams and bridge decks with rapid time by reducing formworks, shuttering and job costs at building sites. Prefabricated buildings are used for places not appropriate for conventional construction techniques, such as the hilly area and where conventional building materials are not accessible readily. Structures which have been in operation on a recurring occasion may be uniform buildings such as mass accommodation, stores, shelters, bus stands, protective cabins, site offices, road crossings, tubular structures, concrete building blocks, etc. [4].

1.2 Prefabrication construction standards

A building consists of all the components of a traditional building that consists of floors, ceilings, walls and pillars, columns and steps, while the main distinction is that the components are created in a regulated setting and taken to the site for assembly, attachment and commissioning. The framework can contain limited quantities of part forms including pillars, columns, frames, roof slabs etc. Components may be used to execute such tasks such as load bearing and enclosure. The weight of the modules should be smaller and therefore quicker and simpler to erect, which will further minimize the injuries at the facility. The components should be ideal for mechanized production or, at any rate, for high mechanization in production. The framework must be constructed according to the structural characteristics such as span, crane weights, illumination, etc. [5].

Prefabricated parts like concrete panels or steel and glass panels need to be treated carefully. The corrosion tolerance of the joints of prefabricated parts must be taken into account in order to prevent the joint loss. The cost of shipping for voluminous prefabricated parts can be higher than for the products from which they are produced, and so often more suitable packaging must be given. Wide prefabricated parts involve heavy-duty cranes, precise measuring and placement [6].

A powerful information processing framework, close to the biological neural network characteristics, is developed by the Artificial Neural Network (ANN). ANNs have a broad variety of highly integrated computing components, either nodes or units or neurons, typically operating in parallel and set up in standard architectures. Each neuron is linked by a link to the other neuron. Each relation link is connected to weights containing input signal information. The neuron net utilises this knowledge to solve a specific issue. Collective behaviour of an ANN is defined by the capacity to understand, remember and generalise training patterns or data identical to the human brain. It may model the original neuron networks as seen in the brain. The computing components for ANN are therefore referred to as neurons or artificial neurons. Figure 1 displays the representation of the statistical model of the artificial neuron [13].

1.png

Figure 1. Artificial neuron mathematical model

Neural networks that can extract importance from complicated or imprecise data can be used to delete patterns and to spot trends that are too difficult to be observed by people or other computer techniques. As an expert in a specific knowledge category, a qualified neural network may respond and evaluate the data. This specialist may be used to include forecasts and solutions to the issue posed in new circumstances of concern. Adaptive learning, self-organization, real time service and fault tolerance through redundant knowledge coding are also advantages of an ANN work. An ANN is a mathematical information processing model which, because of its specific characteristics, is useful and attractive to forecasting tasks: it is a self-adaptive system centred on details, able to learn from past experience; it can generalise the information obtained and accurately infer an unknown part; it has the capacity to estimate a continued feature to any required precision. An ANN is known to be a general simulation method for the estimation of nonlinear time series such as asset markets, foreign exchange rates, accident gravity and high precision traffic volume [14-16]. An ANN was also implemented to estimate the various facets of construction cost across different periods of the life cycle of a building. ANNs and regression approaches have successfully been used with several civil engineering challenges in recent years.

Training involves measures such as evaluate input data performance, equate the estimated performance of each pattern with the goal output of that pattern and change the weights of each neuron to minimize the discrepancy between the amounts that have been tested and estimated, such as measuring the error by repelling an error feature backwards via the neural network. The back-propagation algorithm is used for this training method. One of the most significant improvement in the neural network is the back-propagation learning algorithm. The research and engineering group has been re-energized by this Network to model and process various quantitative phenomena using neural networks. This learning algorithm is applicable to multi-layer feedback networks that consist of processing elements with continuous differentiable functions. The networks connected to the context learning algorithm are often called context propagation networks. This algorithm offers a way to adjust the weights inside a back-propagation network for a specified collection of training input-output pairs to better identify the given input patterns. The basic principle of this algorithm for weight upgrades is essentially the gradient decreasing approach used for simple perception networks of various units. This is a way to spread the error back through the secret machine. The goal of the neural network is to form the network such that it can match the capacity of the net to respond through memorization and its capacity to provide rational responses to inputs close but not equivalent to those that are used through generalisation in training.

The back-propagation algorithm varies from other networks in terms of the mechanism by which weights are determined during the network learning or training phase. The general issue with the multi-component perceptions is to efficiently measure the weights of the covered layers that will contribute to a very slight or zero performance mistake. The network teaching gets more complicated as the secret layers are expanded. The mistake must be determined in order to correct the weights. The error is quickly determined in the output layer, which is the gap between the real or estimated and the expected or target output. It should be remembered that there is no clear information of the mistake on the secret layers. Therefore, other methods for measuring an error on the secret layer can be utilised. This would reduce the production error to a minimal, and this is the overall objective of the current study. The backbone network is trained in three stages such as feedback of the input training pattern, error estimation and context propagation and weight changes. To measure the back-propagation network the feed-forward process is measured only. More than one hidden layer may be more helpful but one hidden layer is plenty. Even if the training is very sluggish, the network will achieve its performance very quickly once educated. The three-layer neural propagation network, which has proven valuable when modelling input-output interactions, are the most frequently utilised communication pattern. The number of neurons in the input and output layers correlates to the sum of input and output variables in the data set while the number of hidden layers or number of neurons for each hidden layer is specified without a particular law. The number of neurons must then be found experimentally in any hidden layer. A network of two secret layers or two sub-layers is usually enough to solve much of the complicated problems in civil engineering applications. A network can be programmed to replicate the desired input-output connection by adjusting weights between the neurons. The nonlinear transition between the input and the output takes place in a concealed layer by using a shift function or activation function, converting the weighted inputs. Linear, sigmoid, log-sigmoid, and tan-sigmoid are the most common activation functions. Neural networks are part of artificial intelligence and have been widely deployed in numerous fields. In view of this strong capability, researchers operated with more strength and accuracy on a new generation of ANNs.

5.1 Training back-propagation network algorithm

In the training algorithm the terminologies used are the following:

x = vector of input training (x₁,….. x_i ..., x_n)

α = learning parameter

t = vector of output with target(t₁, ..., t_k, ......, t_m)

x_i = unit entry i

Because the input layer uses ID activation, here are the exact input and output signs.

w_0j = preference for jth hidden unit w_0k = preference for jth output unit, z_j = hidden unit j.

The net input to z_j is:

$\mathrm{Z}_{\mathrm{inj}}=v_{o j}+\sum_{i, j=1}^{n} x_{i} v_{i j}$         (5)

The net output to z_j is:

$\mathrm{z}_{\text {outj }}=f\left(\mathrm{z}_{\mathrm{inj}}\right)$        (6)

The net input y_k is represented as,

$y_{\mathrm{inp}-k}=\mathcal{w}_{o k}+\sum_{j=1}^{n} z_{j} \mathcal{w}_{i k}$        (7)

And the output is:

$y_{\text {outp-k }}=f\left(\mathrm{z}_{\text {inp-k }}\right)$        (8)

δk = weight change of the error correction for w_jk due to an error in unit y_k that is propagated to hidden units, feeding to unit _ykδj = weight modification of the error correction for v_ij due to back-propagation of the error in the secret device z_j. Furthermore, it should be remembered that binary sigmoidal and bipolar sigmoidal activation functions are widely utilized. Due to these three attributes such as consistency, differentiability and non-decreasing monotony, these functions are included in the back propagation network. The binary sigmoid spectrum is between 0 and 1 and between -1 and 1. The error back-propagation learning algorithm uses the gradual method for weight updates, which immediately adjusts weights when a training pattern is implemented. The following algorithm defines the error back-propagation learning algorithm:

Algorithm-1: Error back-propagation learning algorithm

Step 0: Weight initialization and learning rate initialization.

Step 1: Execute steps 2-9 while stop if incorrect.

Step 2: Execute 3-8 steps with each training pair. Feed forward [Step 1] Feed forward.

Step 3: The input signal x_i is sent to the input device to the s hidden unit i =1 to -n.

Step 4: Each z_j unit secret (j=1 to p) sums up the weighted input signals for calculating the net input: Calculate the performance of the secret device with its activation functions. Z in p-j (function of binary or bipolar sigmoidal activation):

Step 5: Measure the net input for each output unit y_k (k=1 to m).

$y_{\text {inp-k }}=w_{o k}+\sum_{j=1}^{n} z_{j} w_{i k}$

And use the activation functionality to measure the performance signal,

Y_k = f(y_inpk) Error back-propagation [Step 2]

Step 6: Each y_k output unit receives a target pattern matching the input training pattern and calculates the error correction term.

ð_k = (t_k − y_k)

It will measure the f'’(y_ink) derivative from the activation function. Adjust the weight and bias adjustments on the basis of the measured error correction bias.

∆_jk = αð_k z_j

∆w_outputk = αð_k

Also, give δ_k back to the secret layer.

Step 7: Per unit z_j (j=1 to p) secret sums the input units of each unit.

The word δ_inpj is compounded by the f(z_inpj) derivative to determine the error term.

ð_j = ð_inj f(zinj)

The f'’(z_inj) derivative may be determined from activation function based on whether the sigmoidal function is binary or bipolar. Based on the measured ̈δ_j, weight and prejudice modifications are updated.

∆V_jk = αð_k X_j

∆V_outputk = αð_j

Updating weight and prejudice [Step 3].

Step 8: Each yk unit of performance (k=1 to m) updates bias and weight.

w_jk (new) = w_jk (old) + os ∆v_ojk

W_Ok (new) = w_ok (old) + ∆W_ojk

Step 9: Every zj (j=1 to p) secret unit updates its predictions and weight.

v_ij(new) = v_ij (old) + ∆ v_ij

v_Oj(new) = v_Oj (old) +∆ v_oj

Step 10: Search to see if there is a pause. The stop condition could be some periods or whether the real performance compares to the goal performance.

The method for checking the back-propagation network is as follows:

Algorithm-2: Testing the back-propagation network algorithm

Step 0: Weight initialization. The weights are extracted from the algorithm for preparation.

Step 1: Perform steps 2-4 for each vector data.

Step 2: Set the input device activation for xi (i=1 to n).

Step 3: Measure the net input and output of the secret unit x. In j=1 to p,

$\mathrm{Z}_{\mathrm{inj}}=v_{o j}+\sum_{i, j=1}^{n} x_{i} v_{i j}$

$\mathrm{z}_{\text {outj }}=f\left(\mathrm{z}_{\mathrm{inj}}\right)$

Step 4: Measure the performance of the output layer device now. To k=1 to m,

$y_{\text {inp-k }}=w_{o k}+\sum_{j=1}^{n} z_{j} w_{i k}$

$y_{\text {outp-k }}=f\left(\mathrm{Z}_{\mathrm{inp}-k}\right)$

Step 5: To measure the performance, use the sigmoid activation features.