An Ontology for Defect Detection in Metal Additive Manufacturing_2

2025-04-27 0 0 982.62KB 19 页 10玖币

侵权投诉

An Ontology for Defect Detection

in Metal Additive Manufacturing

Massimo Carraturoa, Andrea Mazzullob

aDepartment of Civil Engineering and Architecture, University of Pavia, Italy

bKRDB Research Centre, Faculty of Computer Science, Free University of Bozen-Bolzano, Italy

Abstract

A key challenge for Industry 4.0 applications is to develop control systems for automated manufacturing

services that are capable of addressing both data integration and semantic interoperability issues, as well

as monitoring and decision making tasks. To address such an issue in advanced manufacturing systems,

principled knowledge representation approaches based on formal ontologies have been proposed as a founda-

tion to information management and maintenance in presence of heterogeneous data sources. In addition,

ontologies provide reasoning and querying capabilities to aid domain experts and end users in the context of

constraint validation and decision making. Finally, ontology-based approaches to advanced manufacturing

services can support the explainability and interpretability of the behaviour of monitoring, control, and

simulation systems that are based on black-box machine learning algorithms. In this work, we provide a

novel ontology for the classiﬁcation of process-induced defects known from the metal additive manufactur-

ing literature. Together with a formal representation of the characterising features and sources of defects,

we integrate our knowledge base with state-of-the-art ontologies in the ﬁeld. Our knowledge base aims at

enhancing the modelling capabilities of additive manufacturing ontologies by adding further defect analysis

terminology and diagnostic inference features.

Keywords: Knowledge Representation, Ontologies, Additive Manufacturing, Powder Bed Fusion

1. Introduction

Fully automated manufacturing systems, as well as production-as-a-service frameworks, represent a cor-

nerstone of Industry 4.0 applications [27]. In this context, additive manufacturing (AM), and speciﬁcally

metal additive manufacturing (MAM), is particularly suited to industrial paradigms based on automation,

ﬂexibility, and eﬃciency. Indeed, MAM can be considered as a native digital technology, providing a seam-

less workﬂow from the digital design environment to the ﬁnal product, which can be potentially completed

without any human intervention [30].

However, a broader adoption of MAM technologies in industry is still hindered by such factors as: (i)lack

of widely adopted standardisations and speciﬁcations of material properties, machines, and processes [40];

(ii)lack of adequate digital infrastructures, and interoperability issues between diﬀerent production envi-

ronments [7]; (iii)lack of accessible interfaces providing process information that is easily interpretable by

non-experts [47]; (iv)lack of advanced control systems capable of automatically adjusting, at run-time, the

production parameters [54]; (v)challenges in quality assurance due part accuracy and variability [48].

Thus, achieving semantically transparent and interoperable data sets and systems, to address Points (i),

(ii)and (iii)above, is arguably of paramount importance. In this direction, several approaches based on on-

tology engineering and knowledge representation techniques have been proposed [29, 10, 66, 67, 60]. Broadly

conceived as formal speciﬁcations of conceptualisations over a domain of interest, computational ontologies

(cf. [33] and references therein) have been in particular investigated as a tool to improve interoperability of

additive manufacturing systems that involve human-intensive and domain-expert knowledge management

tasks (cf. Section 2 for a literature survey).

Preprint submitted to Elsevier October 11, 2022

arXiv:2210.04772v1 [cs.AI] 29 Sep 2022

To the best of our knowledge, however, despite the number of domain-speciﬁc ontologies proposed in the

literature to address interoperability issues, less attention has been devoted to another crucial aspect of AM

applications, in particular of Powder Bed Fusion (PBF) MAM: that of defect diagnosis and correction. PBF

is a layer-by-layer process, where a layer of metal powder is spread by means of a roller on top of a build

plate, and metal powder particles are selectively melted by means of a localised moving laser heat source

[26]. At industrial level, PBF MAM is a widespread technology, due to its capability to deliver parts with

high surface quality and remarkable mechanical properties. Nonetheless, the localised nature of the melt

pool induces rapid melting-solidiﬁcation cycles that are responsible for part deﬂections and residual stresses

[9]. Moreover, due to their multi-scale and multi-physics nature, phenomena involved in PBF processes are

diﬃcult to control. Finally, the overall process is characterised by complex and not yet fully understood

relationships among material microstructure, part geometry, process parameters, and mechanical properties

and performances [70]. Such issues can lead to process-induced material discontinuities, e.g. lack-of-fusion

and keyhole porosity, balling, crack and delamination [22].

Even if these process-related defects play a key role inﬂuencing part properties and performances (e.g.,

elastic and elastoplastic behaviour, ultimate tensile stress, and fatigue life), an ontology-based representation

of MAM defects, together with their main properties surveyed in the literature [11], is not yet available.

Such a principled approach, integrating observational data with formalised domain knowledge to determine

the causality links and the complex process-structure-property relationships in manufacturing processes,

represents an important preliminary step for the development of reliable monitoring and control systems

capable of addressing Points (iv)and (v)above [20].

Our contribution aims at ﬁlling this gap, by introducing the novel DefectOnt 1ontology for MAM, specif-

ically PBF-based, defects. This ontology relies on a modular structure, and it aligns with other upper and

domain-speciﬁc ontologies from the literature, to favour development, interoperability and maintenance. It

includes axioms covering the following dimensions: (i)MAM-based categories of defects and related proper-

ties; (ii)spatial notions to express geometrical and topological characteristics; (iii)dimensional characteris-

tics requiring a vocabulary of metrological terms; (iv)sensor-related concepts for observational properties.

DefectOnt is implemented in the OWL 2 [32]2, using the open-source Protégé ontology editor [42]3.

The present article is organised as follow. In Section 2, we discuss related work on MAM defects and

AM ontologies. Then, in Section 3, the design methodology and the development phases of our MAM defect

ontology are described. In Section 4, we illustrate the DefectOnt ontology, presenting in detail the modules

that constitute it. Finally, in Section 5, we discuss future research directions and the main conclusions of

the present work.

2. Related work

2.1. MAM-related literature

Malekipour and El-Mounayri [53] identify, analyze, and classify the most common defects in PBF MAM,

deﬁning the relationships among defects and their contributing parameters. To develop a suitable online

monitoring control strategy, defects are organized into categories based on their manufacturing features and

control purposes. Kyogoku and Ikeshoji [46] review the literature regarding defect generation mechanisms

in PBF processes and their mitigation strategies. Snow et al. [71] outline the state-of-the-art knowledge of

gas porosity and lack-of-fusion ﬂaws due to melt pool instabilities in PBF processes. Grasso and Colosimo

[31] present a defect classiﬁcation of PBF process-induced defects based on process signatures, to support

in-situ monitoring and online defect detection. The present contribution follows defects taxonomies proposed

in [31] (and references therein), integrating it with other literature resources (of both non-ontological and

ontological nature).

1https://github.com/AndreaMazzullo/DefectOnt.

2https://www.w3.org/TR/owl2-overview/.

3https://protege.stanford.edu/.

2.2. Ontology-related literature

Given its closely related focus (despite not overlapping with ours, content-wise), we relied on ExtruOnt, an

ontology for the description of an extruder components proposed by Ramírez-Durán et al. [61], as a gold

standard for the development of our knowledge base, adhering to the authors’ methodological, design, and

presentation choices.

With a broader scope, other upper or domain ontologies for AM have been proposed in the literature. The

Manufacturing’s Semantics Ontology (MASON) [49] is an upper ontology for the conceptualisation of core

additive manufacturing domain notions. The US National Institute of Standards and Technology (NIST)

propose an ontology for AM, that we label NIST AM [80, 81], to support the development of laser and

thermal metamodels. Towards interoperable knowledge and data management in applications, Sanﬁlippo

et al. [66] introduce another ontology for AM (Onto4Additive), based on the upper Descriptive Ontology

for Linguistic and Cognitive Engineering (DOLCE) [13].

Other related AM ontologies include the following (cf. also literature reviews in Sanﬁlippo et al. [66]

and Ramírez-Durán et al. [61]): the Additive Manufacturing Ontology (AMO) [1], based on the upper Basic

Formal Ontology (BFO) [3]; the Innovative Capabilities of Additive Manufacturing (ICAM) ontology [34];

a Smart Applications Reference (SAREF) [23] extension for semantic interoperability in the industry and

manufacturing domain (SAREF4INMA) [18]; the Semantically Integrated Manufacturing Planning Model

(SIMPM) [73]; the Manufacturing Resource Capability Ontology (MaRCO) [37]; the Manufacturing Service

Description Language (MSDL) ontology [2]; the Politecnico di Milano–Production Systems Ontology (P-

PSO) [28]; the Ontology of Standard for the Exchange of Product model data (OntoSTEP) [8]; the ontologies

proposed by Dinar and Rosen [21], Liang [52], Roh et al. [64], and Li et al. [50]. Finally, within the EU-funded

project EMMC, the recently proposed European Materials Modelling Ontology (EMMO) [36] provides a

standard representational upper (nominalistic) ontology framework based on state-of-the-art knowledge on

material modelling and engineering.

3. Ontology design and development

To develop our ontology, we followed the approach proposed by Ramírez-Durán et al. [61], with the adoption

of the NeOn Methodology [77] and in particular of the Six-Phase + Merging Phase Waterfall Ontology

Network Life Cycle Model. This model, allowing for a ﬂexible interplay between pre-existing ontological

and non-ontological resources, consists of the following phases, which will be detailed in the remainder of

this section: initiation, reuse, merging, re-engineering, design, implementation, and maintenance.

3.1. Initiation phase

As by the methodological framework of Suárez-Figueroa et al. [76], we initially developed an Ontology

Requirements Speciﬁcation Document (ORSD), summarised in Table 1, with the following goals: deﬁning

the purpose and the scope of our ontology; selecting the implementation languages; identifying intended

users and uses of the ontology; formulating in natural language groups of competency questions (CQs), to

be expressed and answered by our ontology; providing a pre-glossary of terms appearing in the CQs.

To better illustrate the purpose and the intended uses of our ontology, we present the following simpliﬁed

scenario, involving a MAM production service monitored and regulated by a control system relying, for

instance, on a machine learning architecture. Suppose that, during a 3D printing process, the monitoring

system detects a feature that is classiﬁed (by means of, e.g., a pre-trained convolutional neural network) as

a porosity defect, consisting of a void encapsulated within bulk material. Moreover, assume that the system

controller (based on, e.g., a reinforcement learning mechanism), in an attempt to mitigate the propagation

of the feature, sequentially modiﬁes relevant build chamber parameters, powder handling and deposition

system parameters, and the laser scanning speed, observing that only the latter has an impact on limiting

the defect. The purpose of our ontology is to formalise the domain knowledge required to infer that the

detected porosity is an instance of a process-induced defect, rather than of an equipment-induced one (given

that all other possible equipment-related parameters have been ruled out as inﬂuences on the feature).

In such contexts, our ontology can be used to: provide a (both machine- and human-readable) structured

representation of MAM defects, including their main characteristics and mutual relationships; enrich the

online monitoring and troubleshooting capabilities of controllers by means of logic-based reasoning services;

improve the user interface to MAM production processes, integrating black-box controller systems with a

user-queryable and explainable diagnostic framework.

Finally, we have grouped the CQs along four dimensions, constituting the backbone of our ontology

modular structure: (i)MAM-related (CQAm); (ii)spatial-related (CQSp); (iii)measure-related (CQMe);

and (iv)sensor-related (CQSe).

3.2. Reuse phase

In order to obtain the domain knowledge required to express and formalise relevant properties of MAM

defects, we collect material from both ontological and non-ontological resources. While the latter are based

on (not yet formalised) literature on MAM defects, we relied on already existing additive manufacturing

ontologies to gather structured knowledge related to MAM processes. The selected ontologies were preferred

over other available resources based on the following criteria: (i) possibility of integrating class and property

hierarchies with other non-ontological resources; and (ii) vocabularies capable of expressing properties of

defects determined by the CQs. In the following, we present the relevant material divided along the four

dimensions determined by the groups of CQs.

MAM-related resources As our main non-ontological resources related to MAM defects, we identify: Grasso

and Colosimo [31], providing a classiﬁcation that relates each kind of defect with the main causes

analysed in the literature; Malekipour and El-Mounayri [53], similar to the previous article in scope

and purpose, but providing a diﬀerent taxonomy of defects; Snow et al. [71], focusing mainly on the

internal porosity classiﬁcation. To exploit formal representations of the main notions involved in the

MAM domain, while maintaining a neutral and interoperable environment, we use instead the following

ontologies: NIST AM [81]4; and Onto4additive [66]5. Finally, as a methodological guidance and as

a source for speciﬁc module development (cf. Measure-related resources), we rely on the ExtruOnt

ontology [61]6.

Spatial-related resources To model spatial-related, geometrical or topological, concepts and relations, we

choose GeoSPARQL 1.1 [15]7. Additional spatial-related concepts are inherited from the MASON [49]8

ontology.

Measure-related resources To express metrological features in our ontology, we rely directly on the OM4ExtruOnt

module developed by Ramírez-Durán et al. [61]. This module is obtained by removing all the classes

and properties not relevant to the manufacturing setting from the OM ontology [63]9.

Sensor-related resources As main ontological resource to conceptualise observation- and sensor-related no-

tions, we select the Semantic Sensor Network (SSN) ontology [35]10.

3.3. Merging phase

With the aim of improving semantic interoperability and knowledge exchange in applications, we structure

the upper or domain ontologies identiﬁed in the reuse phase so to be aligned with the DefectOnt framework.

In order to merge these ontologies, we perform the following steps.

4https://github.com/iassouroko/AMontology.

5https://ontohub.org/repositories/additive-manufacturing.

6http://siul02.si.ehu.es/bdi/ontologies/ExtruOnt/docs/.

7https://opengeospatial.github.io/ogc-geosparql/geosparql11/index.html.

8https://sourceforge.net/projects/mason-onto/.

9https://github.com/HajoRijgersberg/OM.

10https://www.w3.org/TR/vocab-ssn/.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

AnOntologyforDefectDetectioninMetalAdditiveManufacturingMassimoCarraturoa,AndreaMazzullobaDepartmentofCivilEngineeringandArchitecture,UniversityofPavia,ItalybKRDBResearchCentre,FacultyofComputerScience,FreeUniversityofBozen-Bolzano,ItalyAbstractAkeychallengeforIndustry4.0applicationsistodevelopcontr...

展开>> 收起<<

An Ontology for Defect Detection in Metal Additive Manufacturing_2.pdf

共19页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

An Ontology for Defect Detection in Metal Additive Manufacturing_2

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: