Digital Manufacturing (OE66)

Unit-3:
Intelligent Manufacturing in Digital Manufacturing Science
Intelligent manufacturing means simulating the intelligent manufacture
activity of human experts through computers by using a highly flexible
and integrated way in every section of manufacturing, and then analyzing,
estimating, concluding, conceiving and determining the manufacturing
problem, with the purpose or replacing or prolonging some part of human
brainwork in the manufacturing environment, and collecting, storing,
perfecting, sharing, inheriting and developing the human experts’
manufacturing intelligence. Its major research contents include intelligent
activity, intelligent machine and the methods of combining these two
things organically, of which the core is intelligent activity.
Intelligent manufacturing is a crucial part of the basic theory system of
digital manufacturing science. It provides the basic theory and method of
implementation for intelligent digital maneuver, intelligent digital design,
intelligent digital machining, intelligent digital control, intelligent digital
process planning, intelligent digital maintenance and diagnosis.
Information fusion and knowledge integration are the main portion of the
intelligent manufacturing system (IMS) which directly affects the quality
and efficiency of system function and product implementation.

3.1 Intelligent Multi Information Sensing and

Fusion in the Manufacturing Process
3.1.1 Intelligent Multi Information Sensing
The core of digital manufacturing is the discretization and digitization of
manufacturing information. The requirement of management, control,
monitoring and diagnosis cannot be satisfied by using a single sensor;
instead, an intelligent multi-sensor system is needed to provide a variety
of data. Intelligent multi information sensing is composed of sensing,
controlling, computing, information processing, network communication,
artificial intelligence and many other technologies.
By using sensors and data acquisition systems, raw data about the
manufacturing process are obtained. There are, in general, three modes of
data acquisition, namely the concentrated acquisition mode, distributed
acquisition and concentrated control mode, and distributed/concentrated
acquisition and concentrated control mode. They are separately shown in
Figs. 3.1, 3.2 and 3.3.
 Fig 3.1: Concentrated acquisition mode

Fig 3.2: Distributed acquisition and concentrated control mode

 Fig 3.3: Distributed and concentrated acquisition and concentrated control

The Application of Sensor in the Processing

Sensor is necessary in manufacturing equipment intelligence. There are
three kind of sensors for applying in machining:
1. Sensor applying in a moving process’s control, for example, the position
sensor, speed sensor, angle speed sensor.
2. Sensor applying in the machining process’s control, for example, the force
sensor, power sensor; and
3. Sensor applying in the machining process’s monitor and diagnosis, for
example, the vibrancy sensor and temperature sensor.

3.1.2 Intelligent Multi Information Fusion

Processing is a very complex dynamic system in the digital manufacturing
system. The control and diagnosis of product quality needs a large
number of state data and information in processing, which includes
strength current, torque, position, speed, acceleration, surface roughness,
temperature, vibration, spacing, chip, acoustic emission and tool wear,
damaged condition and so on. It is difficult to monitor the processing
status with accuracy and integrity with a single sensor, so it needs to
stress intelligent information processing and multi-sensor fusion.
3.1.2.1 General Principles of Multi-Sensor Data Fusion
1. Basic principles of multi-sensor fusion
The fusion of intelligent multi information is an information process which automatically analyzes
the observational information of the multi-sensor using the theory and technology of artificial
intelligence, and optimizes synthetically, to complete the necessary tasks of decision-making and
estimates.

The network structure of Fig. 3.4 is the general pattern of multi-sensor integration and information
integration. Some combined sensors provide information for the system. The first two sensor
outputs X1 and X2 integrate the information in the left lower nodes forming a new information
which is expressed by X1, 2 the third sensor’s output X3 integrates with X1, 2 in the next point that
is, X1, 2, 3 and so on. All the sensor system’s output integrates into a structure in similar manner of
integration. The broken line from the system pane to each node denotes the feasibility of the
information exchange of systems in the process of integration. It should be noted that, before the
integration of sensor 1, 2 we must match the data of each sensor, namely to unionize the data of
each sensor in the space and time coordinate, to make the data belong to the same environmental
position in the same time section.

Fig 3.4: General mode of multi-sensor integration

2. Configuration of the multi-sensor fusion

The partition of the level of information integration is mostly in the following two methods. The first
method is that the integration of information is classified on three levels as low-level (pixels), middle
(features level) and senior (decision?making level). Pixel-level integration is a fuse processing to the
original information of sensors and information on the various stages of the pretreatment,
respectively.

Another method is to classify information fusion into signal-level, evidence?class and dynamic-level,
as shown in Fig. 3.5. Signal-level information fusion is a non-analytical approximate simulation of the
process for which it is difficult to obtain an analytical mathematical model. At this level, the
integration method generally adopts production rules, correlative analysis or trainable artificial
neural network technology. Evidence level requires building a statistical model of the measured
process; the fusion process of evidence level generally obtains some result according to part
reasoning of each sensor firstly, then unites the reasoning, in order to achieve decision-making,
identification, hypothesis testing and control and so on. Dynamic-level information fusion can adopt
the centralized way or the decentralized way. The so-called centralized way first combines the
information of each sensor, then processes the combined information as a whole. In the so-called
decentralized way, each sensor first handles their own information, and then fuses the sensor
information.

Fig 3.5: Thomopoulos hierarchy method

The basic strategy of information fusion is first to fuse the information on the same level, then to
obtain a higher level of information, and import the relevant information at the integration level.
Generally speaking, information fusion is essentially a multi-information integration from bottom to
top level, a process of abstracting information level by level.

3. Process of the multi-sensor fusion

Figure 3.6 shows the whole process of the integration of information. As most of the tested objects
are non-power objects non-power with different characteristics, such as temperature, pressure,
sound, color, and so on, they are first converted into electrical signals, then transformed into digital
signals which can be handled by the computer via the A/D switch. The digital signals require
pretreatment to filter out interference and noise produced by the data collection process, and the
useful signal remained still needs feature extraction.

Fig 3.6: Process of information fusion

Knowledge Engineering in the Life Cycle of a Product
Knowledge engineering is an important part of artificial intelligence, which gets the artificial
intelligence (including repository, knowledge rule, logical reasoning and so on) and CAX (a general
designation of CAD/CAE/CAM/CAPP) system together, to form an integrated system based on the
whole life cycle of the products. Introducing knowledge engineering to the phases of the whole life
cycle of products through the analysis of the corporation issue can help a manufacturing enterprise
to discern, gain, develop, express, decompose, store, deliver and rebirth the knowledge of product
exploitation effectively and quickly, so that the knowledge can be processed, accumulated and
applied.

Knowledge Representation
Knowledge representation is to study the form in which we store the knowledge in the computer in
order to deal with it. The representations frequently used in the field of AI almost always come from
the abstract models by observation and analysis of the intellectual abilities in microscopic and
macroscopic levels by the researcher. According to these theories, we can classify them into three
types as follows. The part class: logic, generation system, semantic network, framework, script,
process and so on; the distribution class: gene, join mechanism; the directness class: kinds of
graphics, images, sounds and artificial environment. In this way, a knowledge representation
architecture tree can be constructed as shown in Fig. 3.7. In this tree, the part representation is fully
researched and is a frequently used method in the field of AI. The distribution representation
method supplies intellectual abilities.

Fig 3.7: Knowledge representation architecture tree

Knowledge Base
The knowledge base system is composed of the instance base, the rules base, the constraints base
and the knowledge base. The knowledge base management system is utilized to add, manage and
maintain the knowledge base. The instance base is used to store a number of successful instances
and new instances generated by the inference engine that is provided by the user. The rules of
knowledge reasoning are stored in a rule base, such as the search of instance, similar algorithm, etc.
The constraints base is used to store many design constraints, such as the geometric constraints,
product performance parameters and other product related regulations and standards. The
knowledge base
is used to store expert experience knowledge, experimental data, design criteria and formulas.
Knowledge data in the knowledge base can be adopted by the inference engine and obtain
knowledge to learn, add or edit through the system.

Structure of Product Knowledge Base

Figure 3.8 describes the basic structure of a product knowledge base. A number of resources such as
the facts right, feature attribute rule, and product data are defined in the knowledge base, which
integrates the structure of the knowledge base into a product model. Resources are integrated into
the entity of objects in the knowledge base. Therefore, to the outside world, the operation of the
knowledge base has become the operation of knowledge objective. As a breakthrough for the
environment of resource restriction in different modes the knowledge base model can be used
freely in other models of resources, thus avoiding a great deal of duplication of definition.

Fig 3.8: Basic structure of product knowledge base

Knowledge Reasoning
In the product life cycle processes, such as product planning, appearance design, engineering design,
and process planning, a wide variety of issues necessitate decision-making through reasoning. The
process from market demand to the design and development, production, sale, use, disposal and
recycling of a product, have to carry out reasoning and decision-making in accordance with the
constraints. The research contents of knowledge reasoning include how to choose knowledge in the
knowledge base, carry out reasoning according to the evidence provided by users, and give answers
to users or complete specific actions.
Knowledge Reasoning in Engineering Design
At present, integration of rule-based reasoning and case-based reasoning, integration of rule-based
reasoning and fuzzy inference rules, integration of case-based reasoning and fuzzy inference rules,
integration of the instance reasoning of artificial neural network and the fuzzy inference of artificial
neural network are widely utilized in KBE systems, such as intelligent CAD, auto parts marking,
intelligent mold design and intelligent CAPP.

Figure 3.9 shows the general process of problem solving of engineering design. In the engineering
design process, the engineers usually do not start with the deductive reasoning (rule-based
reasoning) method, but adopt the analogical reasoning (CBR) model to compare the new design
problem with past design schemes. They then choose the most similar design scheme and modify
the scheme, in order to obtain a new scheme that satisfies the new design requirements. The edit
process can be taken as a deductive reasoning process, that is to say, if a situation emerges, then
certain parameters are modified.

Fig 3.9: General process of project design

Intelligent Manufacturing System

The general definition is: IMS is a man–machine intelligent system that is composed of smart
machines and human experts, which can be a highly flexible and integrated approach using
computer simulation intelligence activities of human expert to analyze, reason, judge, think and
make decisions in the manufacturing process, thus replacing or extending part of the mental work,
at the same time, collecting, storing, improving and sharing the inheritance and development of the
intelligence of human experts. IMS transforms people’s intelligence activities into the intelligence
activities of manufacturing machinery. Figure 3.10 shows the composition of IMS.
The development of IMS is indispensable to artificial intelligence technology (experts technology,
artificial neural networks and fuzzy logic), but different from artificial intelligence. IMS replaces some
of the mental work for research purposes, and requires a system to independently carry out work in
a certain range within the surrounding environment. At the same time, IMS is different from CIMS.
CIMS emphasizes the integration of materials flow of the internal enterprise and the integration of
information flow, but IMS stresses the self-organizing capacity of the entire manufacturing process.
IM completes the manufacturing task in situations of uncertainty and cannot forecast. It has the
following characteristics: self-organizing capacity; self-discipline ability; self-learning and self-
maintenance capabilities, system intelligent integration and so on.

Fig 3.10: Structure of IMS