Geometrical tolerancing and datums

Colin H. Simmons , ... Neil Phelps , in Manual of Engineering Drawing (Fifth Edition), 2020

Datum targets

Surfaces produced by forging casting or sheet metal may be subject to bowing, warping or twisting, and not necessarily be flat. It is therefore impractical to designate an entire surface as a functional datum because accurate and repeatable measurements cannot be made from the entire surface.

In order to define a practical datum plane, selected points or areas are indicated on the drawing. Manufacturing processes and inspection utilizes these combined points or areas as datums.

Datum target symbols

The symbol for a datum target is a circle divided by a horizontal line (see Fig. 22.36). The lower part identifies the datum target. The upper area may be used only for information relating to datum target.

Fig. 22.36. Datum target symbols.

Indication of datum targets

If the datum target is:

(a)

a point, it is indicated by a cross …………… X.

(b)

a line, it is indicated by two crosses connected by a thin line ……………………… X ————X.

(c)

an area, it is indicated by a hatched area surrounded by a thin double dashed chain.

All symbols appear on the drawing view which most clearly shows the relevant surface (see Fig. 22.37).

Fig. 22.37. Indicating datum targets.

Practical application of datum targets

Interpretation – in Fig. 22.38, it is understood that:

Fig. 22.38. Application of datum targets.
(a)

Datum targets A1, A2, and A3 establish Datum A.

(b)

Datum targets B1 and B2 establish Datum B.

(c)

Datum target C1 establishes Datum C.

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780128184820000220

Basics of Geometrical Tolerancing

Georg Henzold , in Geometrical Dimensioning and Tolerancing for Design, Manufacturing and Inspection (Third Edition), 2021

3.3.2 Datum targets

Datum systems can be established from integral features and from restricted areas of integral features; see previous, or from datum targets, see, e.g., Fig. 3.97.

Fig. 3.97

Fig. 3.97. Example of datum targets according to the 3-2-1 method

Datum targets are points (contacting spheres), small areas (contacting planes), or straight lines (contacting cylinders) contacting the part in fixed locations relative to each other or movable in a specified direction. They define an ideal support device for the part. The symbols for datum targets are shown in Table 3.5.

For rigid parts there are two types of datum target systems:

a)

The 3-2-1 method. The primary datum has three datum targets, the secondary two and the tertiary one. See Fig. 3.97. The part does not rock when it rests in the supporting device. The supporting device is shown in Fig. 3.98.

Fig. 3.98

Fig. 3.98. Supporting device according to datum targets of Fig. 3.97

b)

Method of stable primary datum. The primary datum has the minimum number of datum targets in order to lock all its degrees of freedom: in the case of a cylinder, four datum targets. See Fig. 3.102. Six datum targets are used in total. The part does not rock when it rests in the supporting device.

When the contacting features define a datum that is different from the datum feature (e.g. a plane instead a cylinder) the symbol [CF] (meaning contacting feature) shall be indicated; see Figs. 3.102 and 3.106.

The total number of datum targets for rigid parts is always six, in accordance with the six degrees of freedom in a coordinate system (three translations along x, y, z and three rotations around x, y, z). Each datum target removes one of the degrees of freedom. See Fig. 3.100.

Figure 3.98 shows the supporting device according to Fig. 3.97.

Figure 3.99 shows how the datum planes are located relative to the contacting features.

Fig. 3.99

Fig. 3.99. Datum planes located relative to the contacting features

Fig. 3.100 shows the 6 degrees of freedom for a workpiece.

Fig. 3.100

Fig. 3.100. Datum targets, 3-2-1 method; each datum target locks one DOF

Figure 3.101 shows the rules for locating datum targets, in order for each to lock a DOF. E.g. three datum target points for a primary datum must not lie on a straight line. The two secondary datum targets must not lie on a line normal to the primary datum, etc.

Fig. 3.101

Fig. 3.101. Rules for locating datum targets

When the 3-2-1 method is used for a cylindrical datum feature, the primary datum with three datum targets is not yet stable (it becomes stable together with the two datum targets of the secondary datum feature).

Figure 3.102 shows the datum targets to be used (four) for the cylindrical primary datum feature, in order to make it stable.

Fig. 3.102

Fig. 3.102. Primary datum targets on a cylindrical feature to make the datum stable

Figure 3.103 shows the nominal dependency of datum target points or straight lines and the cylinder centre line.

Fig. 3.103

Fig. 3.103. Nominal dependency of datum target points or straight lines and the cylinder centre line.

When on a cylindrical datum feature three or four spherical datum targets are used, the datum plane (start of dimensions) may be defined by the centres or highest or lowest points of the spheres or by the points of nominal contact.

In order to centre the part in the fixture, movable datum targets can be used. Figure 3.104 shows the symbol for the drawing indication.

Fig. 3.104

Fig. 3.104. Symbol for a movable datum target

When three-jaw chucks (movable datum targets) are used to establish a point or part of the axis, the total number of datum targets may be different from 6 datum targets (Fig. 3.105).

Fig. 3.105

Fig. 3.105. Datum target system using a three-jaw chuck

Figures 3.106 to 3.108 show the use of movable datum targets with and without [DV].

Fig. 3.106

Fig. 3.106. Movable datum targets, linear TED with [DF]

Fig. 3.107

Fig. 3.107. Movable datum target, linear TED with [DV]

Fig. 3.108

Fig. 3.108. Movable datum target, linear TED with [DV]

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780128240618000074

Actor's knowledge massive identification in the learning management system

Yassine Benjelloun Touimi , ... Samir Bennani , in Intelligent Systems and Learning Data Analytics in Online Education, 2021

10.2.9 Collect and loading traces process with big data tools

The layer responsible for loading data into big data should be able to handle huge volumes of data, with high speed, and a wide variety of data. This layer should have the ability to validate, clean up, transform, reduce (compress), and integrate the data into the large data stack for processing. Fig. 10.2 illustrates the process and components that must be present in the data load layer.

Figure 10.2. Traces collection and loading process in big data.

The components in the loading and collection process are:

1.

Identification of the various known data formats, by default big data targets the unstructured data.

2.

Filtration and selection of incoming information relevant to the business.

3.

Constant validation and analysis of data.

4.

Noise reduction or removing involves cleaning data.

5.

The transformation can lead to the division, convergence, normalization, or synthesis of the data.

6.

Compression consists of reducing the size of the data, but without losing the relevance of the data.

7.

Integration consists all the data into the big data storage.

Indeed, in this phase the big data system collects massive data from any structure, and from heterogeneous sources by a variety of tools. The data are then stored in the HDFS file format or NOSQL database (Prabhu et al., 2019). In what follows, we will make a comparative study of the tools that make this collection operation with respect to the norms and standards of big data. Then we will look at the different formats for storing structured and unstructured data. The big data collection phase can be divided into two main categories, which depend on the type of load, either batch, microbatch, streaming.

In the big data context, data integration has been extended to unstructured data (sensor data, web logs, social networks, documents). Hadoop uses scripting via MapReduce; Sqoop and Flume also participate in the integration of unstructured data. Thus certain integration tools, including a big data adapter, already exist on the market; this is the case with Talend Enterprise Data Integration–Big Data Edition. To integrate large volumes of data from the building blocks of companies' information systems [enterprise resource planning (ERP), customer relationship management (CRM), Supply Chain (supply chain management)], ETLs, enterprise application integration (systems) (EAIs), and enterprise information integration (EIIs) are always used (Acharjya and Ahmed, 2016; Daniel, 2017).

1.

Batch processing: the big data framework has three modes of data collection. The first mode concerns the collection of massive data done locally then integrated successively in our storage system. The second mode is based on ETL techniques (extract, transform, load). To this end, the system creates a network of nodes for the synchronization of big data. This method responds effectively to the process of extracting and importing big data. A third collection method is the Spooq mode, which allows the step of import/export of data from a relational database to storage based on big data, either in HDFS file format of Hadoop, or to NOSQL tables. This transformation is carried out by the algorithms of MapReduce.

2.

Stream processing: stream loading tools are growing every day, with the appearance of new APIs (application programming interface). This operation can be done in microbatch, and in two modes, either the system is hot or stopped. So we can group these systems into two categories, either in real time, that is, hot extraction, even if the system is in full production, or with batch and microbatch, in a well-determined time, for small amounts of data, which requires shutting down the production system.

All of the above features are implemented by the tools of the big data framework (Acharjya and Ahmed, 2016):

1.

Apache Flame interacts with log data by collecting, aggregating, and moving a large amount of data with great robustness and fault tolerance.

2.

Apache Chukwa is based on the collection of large-scale system registers. It does the storage by exploiting NoSQL or HDFS storage.

3.

Apache Spark is a framework which has as a basic principle, the treatment of all types of big data. Spark integrates streaming processing with the Spark streaming API. The latter allows microbatch processing and storage of all massive data to anyone with a big data storage system with RDD support (resilient distributed dataset).

4.

Apache Kafka is a data collection platform, which offers three methods of batch streaming: the stream API makes it possible to collect massive streaming data and stores it in topics, the producer API publishes streaming data for one or more subjects, the connector API—this is an API for application access connections.

After citing most of the tools from the big data framework that provide the different data collection modes of heterogeneous platforms, we will detail in the next section the data processing phase.

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780128234105000097

Metabolic Engineering of Enzyme-Regulated Bioprocesses

Anshula Sharma , ... Baljinder Kaur , in Advances in Enzyme Technology, 2019

11.3.3.2 Comprehensive Omics Data Analyses

Analyses of omics, including genomics, transcriptomics, proteomics, metabolomics, and fluxomics, can provide invaluable system-wide information on cellular and metabolic characteristics under various genotypic and environmental conditions. By analyzing omics data, target genes to be manipulated can be identified for enhancing chemical production capability. Recently, interest in fluxomics has been increasing, because the ultimate purpose of metabolic engineering is to optimize flux distribution toward product formation. The multiomics approach can offer a comprehensive understanding of the host strain by compensating each omics drawback. The multiomics approach can be used to elucidate various phenomena in a metabolically engineered strain, and to identify further engineering targets. For example, by analyzing metabolomic and fluxomic data, low glycolysis flux was revealed to be one of the potential reasons for inefficient ethanol production by S. cerevisiae when using xylose as a carbon source [39]. More recently, transcriptomics and proteomics were applied simultaneously to reveal an underlying mechanism for sucrose and glycerol dual carbon source utilization of homosuccinic acid producing a Mannheimia succiniciproducens PALFK strain. The analysis showed that deletion of the fruA gene encoding fructose phosphotransferase system (PTS) system allowed the simultaneous uptake of sucrose and glycerol by increasing the cAMP level, which activates the glycerol utilization pathway [40]. Multiomics approaches are not only applied to metabolically engineered strains, but also to wild-type host strains to select the best potential host strain for further metabolic engineering.

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B978044464114400011X

Support Vector Machine: Principles, Parameters, and Applications

Raoof Gholami , Nikoo Fakhari , in Handbook of Neural Computation, 2017

27.5 Step by Step with SVMs for Classification and Regression Data Analysis

Having known the principles and equations of the SVM, the following steps can be followed to have a successful classification or regression tasks:

(i)

Preparation of a pattern matrix: The pattern (feature) matrix required for the classification and regression analysis is different. The one used for the classification should have a set of data divided into two or multiple classes of extracted features with a signature such as 1 or −1, indicating the class from which that feature belongs. For the regression analysis, on the other hand, independent (input parameters) and dependent data (target parameters) are often presented in separated columns of the matrix. Data can then be further partitioned into training, testing, and validation portions.

(ii)

Selection of a Kernel function: This is perhaps the most important step. There are many kernel functions which might be applied but their applications depend, to a great extent, on the nature of data (i.e., the degree of nonlinearity). The RBF (Gaussian) kernel function has been the most successful kernel based on its application reported in the literature [5,23] and can be considered as the first choice.

(iii)

Parameter selection: When SVMs are used, there are a number of parameters selected to have the best performance including: (1) parameters included in the kernel functions, (2) the trade-off parameter C, and (3) the ε-insensitivity parameter. Selection of these parameters, however, is not an easy and straightforward task as there are no mathematical equations or correlations to give an initial guess of their values. Values suggested by famous softwares (i.e., Weka) can be used/modified under these circumstances or a validation step can be set apart to determine the values of the parameters.

(iv)

Execution of training algorithms: When input and output data are defined at the training step, an SVM uses the general formulations presented earlier as Eq. (27.8), Eq. (27.16), Eq. (27.19) Eq. (27.27) or Eq. (27.30) to determine the Lagrange multipliers. The multipliers with nonzero values would indicate which one of the input data ( x i ) can be a support vector. The support vectors would determine the margin of each class through which the optimum hyperplane (decision function) can be chosen.

(v)

Classification/prediction of unseen data: By determination of Lagrange multipliers and their corresponding support vectors, unseen data can be properly classified or estimated. Having failed in having a reliable classification or prediction tasks at this stage might be due to a bad feature extraction/selection, kernel selection or parameters estimation at any of the steps explained above. One may repeat the above steps to reduce the error and enhance the accuracy of the results under these circumstances.

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B9780128113189000272

22nd European Symposium on Computer Aided Process Engineering

Mona Gharaie , ... Nan Zhang , in Computer Aided Chemical Engineering, 2012

1 Hierarchical Approach

A typical total site includes several plants (or 'processes') converting raw materials to desired products. Given CO2 emissions reduction target for a site, the most appropriate level of energy saving depends on carbon emissions penalties (or tax). A four-step hierarchical method for assessing CO2 emissions reduction options is introduced. The first step is to modify the processes, e.g. changing catalyst, introducing new technologies, changing operating conditions, etc., in order to reduce heat demand and/or improve opportunities for heat recovery. In the second step, HEN retrofit increases heat recovery and hence reduces demand for external utilities from the utility system and/or local furnaces. In the third step, operational optimization of the utility system is carried out. Finally, fuel switching, i.e. selection of a lower carbon-content fuel for burning in boilers and furnaces to supply external utilities, is considered. Since carbon trading opportunities (or other penalties or taxes) depend on other carbon emissions reduction activities, carbon trading is included in the problem analysis. Figures 1 and 2 represent the proposed hierarchy and calculation strategy, respectively, for CO2 emissions reduction in a total site. Note that process changes are not considered in this paper in any detail.

Fig. 1. Hierarchy for site CO2 Emission Reduction

Fig. 2. Hierarchical retrofit design method computational procedure

Table 1 presents a simple problem to illustrate the hierarchical design methodology to find a cost effective combination of CO2 emissions reduction strategies. The available capital investment is assumed to be 10 MM$. The emissions reduction target is 126.95 kt/yr. The payback time limit is 2 years. It is assumed that all practical process changes have been made. The site includes three process units and a utility system. Furnaces supply heat to process units A and C while the utility system provides medium pressure steam to meet the heating requirements of process B. The total electricity demand of the process units is 5.7 MW. Fuel data are shown in Table 2.

Table 1. Case study data

Process Unit Type of Utility CO2 Emissions (kt/yr) Hot Utill. Demand (kW)
A Furnace 319.37 80418
B Boiler 122.9 10190
C Furnace 74.66 18800

Table 2. Fuel data

Fuel Type NHV (kt/kg) Carbon (wt%)
Syngas 12309 37
Coal 29000 70
NG 52206 74

Process stream data (target temperature, supply temperature, enthalpy change) for Process C are shown in Table 3. The existing HEN structure of Process C is shown in Figure 3, as an example.

Table 3. Process stream data of Process C

Steam TSupply (°C) TTarget (oC) Flow rate (kW/oC) Stream TSupply (oC) TTarget (oC) Flow rate (kW/oC)
Hot 1 165.5 90 22.6 Hot 8 204 104 5.4
Hot 2 282 196.5 54.4 Hot 9 140.9 38 162
Hot 3 274 37.5 9.1 Hot 10 144.5 51 15.2
Hot 4 164 27 36.1 Cold 11 74 295 62.4
Hot 5 327 261 44.3 Cold 12 143 164 1293
Hot 6 363 246 26.7 Cold 13 94 125 126.4
Hot 7 327 165 16.7

Note: Exchanger additional area capital cost ($) =9,665(area) 0.68, New exchanger unit capital cost ($) =94,093+1,127(area)0.98, Hot utility cost ($/kWy) =306.8, Cold utility cost ($/kWy) =5.25

Fig. 3. Grid diagram representing existing heat exchanger network of Process C

The utility costs are straightforward to calculate given the demand for each utility. The unit cost of utilities and correlations to calculate heat exchanger capital cost from areas are taken from Smith et al., 2010 and presented in Table 3.

Step 1: Modifications for retrofit of an existing HEN include structural changes (e.g. installation of a new heat exchanger, splitting streams, re-piping and re-sequencing existing piping) and adding area to existing heat exchangers. The network pinch method (Asante & Zhu, 1997; Smith et al., 2010), implemented in SPRINT software, optimizes the HEN taking capital-energy trade-offs into account. Table 4 presents some possible modifications and the associated CO2 emissions reduction.

Table 4. Retrofit modifications of HEN of Process C

Process Unit ID Modification Additional Area (m2) Energy Saving (k$/y) Investment Cost (k$) CO2 Emissions Reducation (t/y)
C 1 Add. Area 2,499.48 1,808 2,766.09 34,948.63
2 Add. Area 3,528.83 2,250 3,497.18 43,497.10
3 Add. Area 4,674.66 2,780 4,234.08 53,753.3.0
4 Add. Area 6,028.54 3,311 5,033.54 64,009.53
5 Add. Area 9,062.81 3,862 6,641.53 74,662.98

In order to assess the HEN retrofit options, a parameter, (ξ = ΔMCO2/IC), is defined to represent the effectiveness of the option in terms of CO2 emissions reduction (ΔMCO2 ) with respect to investment cost (IC). Given the target for CO2 emissions reduction and the capital budget, the break-even cost-effectiveness (ΔCO2 total/IC total) can be used to screen HEN retrofit options, as illustrated in Fig. 4. The shaded region in Fig.4 represents HEN retrofit options that are at least as effective as the break-even measure. Combinations of HEN retrofit options within the capital budget (10 MM$), are determined by searching among the cost-effective options. Table 5 presents some combinations of HEN retrofit options within the highlighted region requiring 10 MM$ and the related CO2 emissions reduction

Fig. 4. Retrofit profiles of process unit HEN

Table 5. Cost-effective HEN retrofit options

Option Process Investment Cost (MM$) Total CO2 Reduction (kt/yr)
I A 9.09 147,225
C 0.91
II A 8 152104
C 2
III A 6.19 152,392
C 3.81

Step 2: Operational optimization of the utility system is explored, given the utility demand after HEN retrofit. The optimization considers trade-offs between power generation and fuel consumption for minimum operating cost taking into account carbon tax (Varbanov et al., 2004). Analysis of the utility system is carried out using Mixed Integer Non Linear Programming (MINLP) optimization in STAR software. Table 6 presents the performance of utility system after operational optimization.

Table 6. Performance of utility system

Base Case Optimized
Power Generation (kW) 6039.3 6578.3
Total Oper. Cost (MM$/y) 10.9 9.4
CO2 Emissions (kt/y) 122.9 106.4

Step 3: In fuel switching analysis, the CO2 emissions reduction is calculated from the calorific value and carbon content of the two fuels, the process heat demand, and the efficiency of the furnace or boiler (Delaby, 1993). The relative fuel prices affect the economics of fuel switching. In this study, the price of syngas burning in local furnaces is similar to that of natural gas, the cleaner fuel. Table 7 summarizes fuel switching opportunities.

Table 7. Fuel switching opportunities

Scenario Fuel Cons. (kt/yr) CO2 Reduc. (kt/yr)
I 42.85 130.42
II 42.00 127.84
III 41.95 127.69

The hierarchical procedure is applied to explore options iteratively to identify the most promising combination of options is achieved. Table 8 presents the combined effect of the three steps. User interaction is required to generate solutions, which is potentially time consuming. Moreover the quality of the solution is highly dependent on the design sequence, so is unlikely to lead to an optimal solution.

Table 8. CO2 emissions reduction in hierarchical design method

Base Case HEN Retrofit Utility Optimization Fuel Switching
Utility System CO2 Emissions (kt/y) 122.9 - 106.4 -
Process Units CO2 Emissions (kt/y) 394 238.5 - 113
Total Site CO2 Emissions (kt/y) 516.9 238.5 344.9 219.4
CO2 Reducation (kt/yr) - 155.5 172 297.4

Read full chapter

URL:

https://www.sciencedirect.com/science/article/pii/B978044459519550071X

Critical evaluation of cognitive analysis techniques for construction field management

Fernando A. Mondragon Solis , William J. O'Brien , in Automation in Construction, 2014

3.1 Categorization of the main aspects of cognitive analysis techniques

The sequence of analysis begins by identifying the aspects of cognitive analysis that are relevant to determine a technique's capabilities of analysis and possible applications. Crandall et al. [10] recognize three primary aspects of cognitive analysis that are considered critical to the success and completeness of a cognitive analysis study. These aspects are: (1) knowledge elicitation, which refers to the collection of information about what practitioners know about their work, and how they know it; (2) data analysis, which is the process structuring data, identifying findings, and discovering their meaning; and (3) knowledge representation, which focuses on the tasks of displaying data, presenting findings, and communicating meaning. Crandall et al. [10] also present other categories for cognitive analysis techniques, focused on where to look for data, how to look for it and how to represent it. These categories include four types of data collection methods: interviews, self reports, observations and automated capture; four types of data targets: time, realism, difficulty, generality, and four types of analytical products: textual descriptions, tables/graphs/illustrations, qualitative models, and simulation and symbolic models. In this paper, each category is associated to one of the three main aspects of analysis, in order to analyze the capabilities and applications of techniques: data collection methods are paired to knowledge elicitation, data targets to data analysis and analytical products to knowledge representations. The resulting associations are shown in Table 4, which lists the categories corresponding to each aspect as well as a description of each category.

Table 4. Mapping of aspects of cognitive analysis methods and categories found in Crandall et al. [10].

Aspects and categories Description
Knowledge elicitation methods
Interview Structured set of questions asked by analyst
Self-reports Have practitioners talk about or record their behavior/strategies
Observation Analyst takes notes of practitioners' activities
Automated capture Use of computers to record information and decisions

Focus of data analysis
Time Analyzed tasks are part of present, past or future activity
Realism Tasks are taken from real world activity, hypothetical scenarios or artificial environments
Difficulty Tasks are part of routine/typical events, challenging events, or rare events/anomalies
Generality Whether the purpose is to analyze abstract/general knowledge, a job/task, or a specific incident/event

Products of knowledge representation
Textual descriptions Analytic products are mostly text based
Tables, graphs, illustrations Analytic products include organized text in tables, graphs, drawings and other types of illustrations to represent the data
Qualitative models Flow charts and diagrams that describe knowledge
Simulations, models Knowledge is represented through an entire numerical, or symbolic model, including computer models

This classification is then used to understand the applicability and suitability of cognitive analysis techniques. This analysis begins by defining the type of problem a technique can address through the categories of data analysis. To exemplify which aspects have been successfully applied for construction research, the techniques reviewed in the previous section are broken down for their data analysis capabilities. Then, the outcome of each technique is analyzed through the categories of knowledge representation. These categories are helpful to understand what each technique can communicate and its potential applications, and this also gives insight into the resources that are necessary to develop the products of each technique. The knowledge representations of the techniques reviewed in Section 2 are used to analyze which applications can be supported by each technique. Finally, the categories of knowledge elicitation give further insight into the resources that the analyst and the subject of analysis must provide in order to carry out the cognitive study. Altogether, this analysis serves as a guide for selecting techniques from the large number found in the literature. Furthermore, this analysis also enables the discussion of the practical implications of techniques in the context of construction jobsites, as it lays out detailed characteristics of each of the three aspects of cognitive analysis. This discussion is presented in Section 4.

Read full article

URL:

https://www.sciencedirect.com/science/article/pii/S0926580513002331

Recent developments and trends in point set registration methods

Baraka Maiseli , ... Huijun Gao , in Journal of Visual Communication and Image Representation, 2017

2.2.6 Coherent Point Drift (CPD) algorithm

In [83,17], Myronenko et al. proposed a probabilistic approach to register point sets under both rigid and non-rigid transformations. Their method, called Coherent Point Drift or CPD, estimates the probability density function that can optimally align two point sets. The authors iteratively fitted the GMM centroids of the source point set to the (data) target point set by maximizing the likelihood and finding the centroids' posterior probabilities. Next, to maintain the topological structure and configurations of the point sets, the centroids were forcefully moved coherently as a group. The CPD algorithm, which exhibits a linear computational complexity, outperforms most state-of-the-art algorithms and achieves promising results under conditions of noise, outliers, and missing points.

Despite being popular and superior, studies have revealed weaknesses of the CPD algorithm. In [84], Wang et al. noted from the CPD model a weight parameter w that predicts the level of noise and number of outliers in the GMM. This value was manually selected under unclear constraints, hence raising concerns regarding reliability of the registration results. Wang et al., therefore, introduced an algorithm to automatically select an optimal value of w. Their algorithm is based on the hybrid optimization scheme, which integrates the generic and the Nelder-Mead simplex algorithms [85–88]. Through quantitative evaluations, the modified CPD algorithm was observed to produce more accurate results compared to its competitor. Recent improvements of the CPD algorithm are in [89–98].

Read full article

URL:

https://www.sciencedirect.com/science/article/pii/S1047320317300743

Review of predicting the blast-induced ground vibrations to reduce impacts on ambient urban communities

Yu Yan , ... Honglu Fei , in Journal of Cleaner Production, 2020

3.2.3 Classification and regression tree (CART)

Classification and regression tree (CART) is a kind of decision tree, which is composed of feature selection, tree generation and pruning. CART is a binary tree in which the inputs of the system represent the parameters to determine the results of model outputs. As illustrated in Fig. 24, the structure of a CART tree generally consists of a root node, several operational nodes and several leaf nodes.

Fig. 24

Fig. 24. Structure of a simple CART model.

CART models can effectively solve the non-linear problem of data samples without assuming samples, and hence, are suitable for solving the complex problem of unclear relationship between data samples and target variables. The key of CART models is the proper selection of the optimal partition attributes, such that samples contained in the branch nodes as divided by CART models belong to the same class as much as possible. CART models use Gini index to classify the attributes of root nodes, and they stop building trees until each sample is completely pure after partitioning. Then the purity of data set can be measured by the Gini index as.

(22) Gini ( D ) = 1 − ∑ i = 1 N p i 2

where, N is the number of class labels, and p i is the proportion of i samples in the total samples. The smaller the Gini index, the higher is the purity of data set D. The over-fitting of CART model is solved by pruning, and the pruning is further divided into pre-pruning and post-pruning.

CART is considered as an interesting method for regression analysis in numerous areas, especially in blast engineering. Khandelwal et al. (2017) adopted the CART model for forecasting the PPV by considering Q and R, and concluded that the CART model exhibited higher prediction capability when compared to empirical equations and multiple regression model. Hasanipanah et al. (2017) established the CART model for predicting the PPV by choosing a different parameter of root node in contrast to Khandelwal et al. (2017), and concluded that a CART model with the R2 of 0.95 predicted the PPV better than the empirical models.

To simplify the CART modeling, maximum tree depth and number of intervals were appropriately selected so as to limit the excessive growth of the tree and the existence of overfitting problems. To improve the accuracy of predicting the PPV, maximum tree depths and numbers of intervals were selected by trial and error in the literature (Khandelwal et al., 2017; Hasanipanah et al., 2017). The datasets used in the two references were 51and 70 respectively, and the values of statistical criteria showed relatively better performance of predicting the PPV using the CART algorithm.

Read full article

URL:

https://www.sciencedirect.com/science/article/pii/S0959652620311823

Rate management in multiuser detection based MAC design for ad hoc networks

Pegdwindé Justin Kouraogo , ... François Gagnon , in Ad Hoc Networks, 2013

6 Conclusion

We analysed the performance of the multiuser detection based MAC protocol interacting with multiuser physical layer. The MUD-based protocol used alone increases users' throughput. However more performance gains are achievable when creating synergy between physical and MAC layer. For this purpose we proposed a predictive framework that relies on the users control packets and the channel gain prediction for the data transmission slot of each user. This prediction is used for the link rates adaptation. At the same time four kinds of detection filters were studied.

The results indicate that channel prediction at MAC layer induces significant performance improvement in the MUD-based ad hoc network. This has been highlighted by comparing the performance of our predictive framework with the estimation based approach and the perfect information case. In particular, the prediction case gives about three times better goodput than the estimation and about 70% of the perfect case goodput. Moreover, comparison with single rate transmission cases, i.e. without rate adaptation shows that the prediction case gives 10-fold increase of the goodput. It is also important that the proposed approach ensures more flexibility in terms of data rates and target packets error rates. At the detector level, the MMSE and decorrelator guarantee high quality reception and their throughput doubles that of the SIC receiver which can ensure only medium quality packets reception. Concerning matched filter reception, it can ensure only low quality packet reception. At the transmitter level, our study shows that multi-code and multi-processing gain give significantly better performances than multi-modulation in terms of goodput and PDR. The choice between the MPG, MC and MM schemes should be guided by implementation issues of the station in the MUD-based ad hoc network, for example the MC has to be implemented with a linear amplifier. In QoS guarantee mode, which is of major importance for multi-media applications, we studied the prediction framework under the constraint of scheduling only the packet transmissions that will meet a required probability loss. The results show the superiority of the proposed PER scheduling, which more than doubles the goodput when compared to the no adaptation case.

As indicated above, the channel prediction-based rate adaptation achieves promising results. However, the fading effect on the system is difficult to combat. The performance could be improved by implementing diversity techniques at the receiver to increase the received SNR. Also, the system is tested here on a Rayleigh channel in a simple scenario where the receiver is fixed and transmitters are in motion and it would be interesting to analyse the system behaviour in a mobile-to-mobile channel. At the scheduling level, instead of the simple PER scheduling scheme, significant throughput gain could be achieved with multiuser detection based schedulers, developed recently, that could takes advantage of the MUD-based protocol structure and the multiuser detection physical layer structure. These topics we will be treated in the future research.

Read full article

URL:

https://www.sciencedirect.com/science/article/pii/S1570870512000649