Professional Documents
Culture Documents
Proceedings of The Third International Scientific Conference "Intelligent Information Technologies For Industry" (IITI'18) : Volume 1 Ajith Abraham
Proceedings of The Third International Scientific Conference "Intelligent Information Technologies For Industry" (IITI'18) : Volume 1 Ajith Abraham
https://textbookfull.com/product/proceedings-of-the-third-
international-scientific-conference-intelligent-information-
technologies-for-industry-iiti18-volume-2-ajith-abraham/
https://textbookfull.com/product/emerging-technologies-in-data-
mining-and-information-security-proceedings-of-
iemis-2018-volume-1-ajith-abraham/
https://textbookfull.com/product/proceedings-of-the-fourth-
international-scientific-conference-intelligent-information-
technologies-for-industry-iiti-19-sergey-kovalev/
https://textbookfull.com/product/emerging-technologies-in-data-
mining-and-information-security-proceedings-of-
iemis-2018-volume-2-ajith-abraham/
The ShortTube 80 Telescope A User s Guide Neil T.
English
https://textbookfull.com/product/the-shorttube-80-telescope-a-
user-s-guide-neil-t-english/
https://textbookfull.com/product/essentials-of-hypertension-
the-120-80-paradigm-1st-edition-flavio-danni-fuchs-auth/
https://textbookfull.com/product/intermittent-fasting-for-women-
over-50-80-recipes-included-1st-edition-sharon-milson/
https://textbookfull.com/product/sew-your-own-wardrobe-more-
than-80-techniques-1st-edition-alison-smith/
https://textbookfull.com/product/unpublished-fragments-from-the-
period-of-dawn-winter-1879-80-spring-1881-1st-edition-friedrich-
nietzsche/
Advances in Intelligent Systems and Computing 874
Ajith Abraham
Sergey Kovalev
Valery Tarassov
Vaclav Snasel
Andrey Sukhanov Editors
Proceedings of the
Third International
Scientific Conference
“Intelligent Information
Technologies for
Industry” (IITI’18)
Volume 1
Advances in Intelligent Systems and Computing
Volume 874
Series editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: kacprzyk@ibspan.waw.pl
The series “Advances in Intelligent Systems and Computing” contains publications on theory,
applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all
disciplines such as engineering, natural sciences, computer and information science, ICT, economics,
business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the
areas of modern intelligent systems and computing such as: computational intelligence, soft computing
including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms,
social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and
society, cognitive science and systems, Perception and Vision, DNA and immune based systems,
self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics including
human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent
data analysis, knowledge management, intelligent agents, intelligent decision making and support,
intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings
of important conferences, symposia and congresses. They cover significant recent developments in the
field, both of a foundational and applicable character. An important characteristic feature of the series is
the short publication time and world-wide distribution. This permits a rapid and broad dissemination of
research results.
Advisory Board
Chairman
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
e-mail: nikhil@isical.ac.in
Members
Rafael Bello Perez, Universidad Central “Marta Abreu” de Las Villas, Santa Clara, Cuba
e-mail: rbellop@uclv.edu.cu
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
e-mail: escorchado@usal.es
Hani Hagras, University of Essex, Colchester, UK
e-mail: hani@essex.ac.uk
László T. Kóczy, Széchenyi István University, Győr, Hungary
e-mail: koczy@sze.hu
Vladik Kreinovich, University of Texas at El Paso, El Paso, USA
e-mail: vladik@utep.edu
Chin-Teng Lin, National Chiao Tung University, Hsinchu, Taiwan
e-mail: ctlin@mail.nctu.edu.tw
Jie Lu, University of Technology, Sydney, Australia
e-mail: Jie.Lu@uts.edu.au
Patricia Melin, Tijuana Institute of Technology, Tijuana, Mexico
e-mail: epmelin@hafsamx.org
Nadia Nedjah, State University of Rio de Janeiro, Rio de Janeiro, Brazil
e-mail: nadia@eng.uerj.br
Ngoc Thanh Nguyen, Wroclaw University of Technology, Wroclaw, Poland
e-mail: Ngoc-Thanh.Nguyen@pwr.edu.pl
Jun Wang, The Chinese University of Hong Kong, Shatin, Hong Kong
e-mail: jwang@mae.cuhk.edu.hk
Andrey Sukhanov
Editors
123
Editors
Ajith Abraham Vaclav Snasel
Scientific Network for Innovation VSB-Technical University of Ostrava
and Research Excellence Ostrava, Czech Republic
Machine Intelligence Research Labs
(MIR Labs) Andrey Sukhanov
Auburn, WA, USA Rostov State Transport University
Rostov-on-Don, Russia
Sergey Kovalev
Rostov State Transport University
Rostov-on-Don, Russia
Valery Tarassov
Bauman Moscow State Technical University
Moscow, Russia
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
v
Organization
Organizing Institutes
Conference Chairs
Conference Vice-chair
vii
viii Organization
Organizing Vice-chair
Invited Papers
Connected Vehicle Prognostics Framework for Dynamic Systems . . . . . 3
Omar Makke and Oleg Gusikhin
Human-Computer Cloud for Decision Support: Main Ontological
Models and Dynamic Resource Network Configuration . . . . . . . . . . . . . 16
Alexander Smirnov, Tatiana Levashova, Nikolay Shilov,
and Andrew Ponomarev
Enterprise Total Agentification as a Way to Industry 4.0:
Forming Artificial Societies via Goal-Resource Networks . . . . . . . . . . . . 26
Valery B. Tarassov
xi
xii Contents
1 Introduction
Vehicle prognostics and health management has always been a topic of great interest
for the automotive industry. It is one of the early areas of intelligent systems appli-
cations targeting on-board and service bay diagnostics and prognostics [1]. In recent
years, due to increased complexity of the vehicle systems and the emergence of con-
nected and autonomous vehicles, the interest in prognostics and health management has
soared.
Connected vehicle technology brings game changing opportunities to advance
vehicle prognostics and health management [2]. Connected vehicles can take advantage
of cloud resources to build complex models, and they can leverage data from multiple
vehicles and other external sources to infer complex relationships and offer enhanced
user experience. Even existing on-board monitoring and diagnostics systems can be
enhanced using connected vehicle technology. For example, modern vehicles are
equipped with the oil-life monitoring system that tracks oil wear by analyzing duty-
cycle and typically provide advanced notice to the driver when the oil life is at 5%.
However, for scheduling purposes, it is more convenient to estimate the “time to oil
change” rather than the oil wear percentage. By using connected vehicle data, it is
possible to project the time to oil wear, assuming consistent duty cycle. Then, a
notification containing projected time to service can be provided to the customer.
Extending the service need prediction to other components, such as brake pads,
tires, batteries, air filter, etc. will provide a comprehensive service-scheduling package,
which offers a new level of user experience and service efficiency. This capability can
insure that the dealer has the right replacement parts available. It can also improve the
efficiency of service parts inventory. For fleet owners, it facilitates better scheduling to
minimize the impact of vehicle downtime on the fleet operations.
Connected vehicle analytics has a promise to develop such models utilizing exiting
vehicle data to project the time to service without adding costly sensors or other
hardware. However, practical implementation of connected vehicle prognostics faces a
number of challenges such as on-board sensor data, memory availability, and limita-
tions of the wireless communication bandwidth [3]. These limitations result in potential
data loss of data which may be crucial for adequate prognostics of the given compo-
nent. To overcome these challenges, the design in [4] proposed the model for con-
nected vehicle brake prognostics, where an on-board system implements a duty-cycle
aggregation model to alleviate limitations of communication channel while a cloud
based model uses cross-vehicle data in lieu of the lack of direct measurement of brake
wear. In this paper we generalize and formalize the approach presented in [4] by
providing a theoretical framework to build such prognostics models for the components
which can be modelled as dynamic systems.
This paper is organized as follows. Section 2 provides an overview of the proposed
framework. Section 3 illustrates the framework using brake pad prognostics as an
example. Section 4 presents the detailed mathematical foundation behind the frame-
work. Section 5 demonstrates how the same framework can be applied to a different
domain area, such as the cabin air filter prognostics. Section 6 concludes the paper and
discusses future work.
The first is physics based prognostics. In this approach, elaborate physics models
are needed with details about many of the components material properties, such as
coefficient of friction, mass, heat capacity, and other material properties. The second
approach is a data-driven approach which is based solely on data without any
Connected Vehicle Prognostics Framework for Dynamic Systems 5
consideration to the underlying physics. The goal here is either to use machine learning
or other statistical tools in order to identify patterns in the data and make projections on
when to service a component. In our framework, we will utilize a hybrid approach
where an in-vehicle physics-based model generates data in a meaningful way, and this
data is sent periodically to the cloud to make predictions using data-driven approaches,
which feeds back a wear result to the vehicle. These in-vehicle models should be
designed to be embeddable (in particular, through over the air updates) into vehicle
control modules, such as modem, engine controller, brake controller, etc.
The problem can be formulated as follows. Suppose that the dynamic behavior of a
component can be described with a physics-based dynamic model, requiring a vector
u of inputs which affects a state vector x. The wear of the component is assumed to be a
function of the inputs and the states. The goal is to find a prognostics algorithm which
outputs “expected time to service” by calculating the “wear” of the component, and
based on the past history, predicts a time to “service component by”.
To minimize the reliance on multiple parameters related to material properties and
environment, machine learning is used. The algorithm must run in the vehicle within
deterministic time period, must have resistance to connectivity issues, must require
fixed size of storage space in the vehicle, and must consume a deterministic amount of
data when transferring the data from the vehicle to the cloud. The framework which
implements such algorithms with these requirements is shown in Fig. 2.
Using this approach, a fleet manager with a known number of vehicles can pur-
chase a data plan adequate enough to support the data transfer, or can ensure that the
fleet vehicles are connected to the cloud at least once every specified period of time,
such as day, a week, or similar. The vehicle generates data based on duty cycle, and the
cloud utilizes this data in machine learning. As will be shown in the following sections,
the method of generating data is an important factor in combining the physics approach
with machine learning.
6 O. Makke and O. Gusikhin
The main objective of the brake pad prognostics system is to give customers a close
approximation of the date when they need to change their brake pads. The algorithm
described here will serve as an example to help in understanding the general framework
for prognostics explained in the next section. The classical approach to this problem is
to create an elaborate physics model which relies heavily on material properties of the
pads and execute it in the vehicle [6]. One simple form of prognostics is a distance-
based prediction. A driver is asked to inspect the brake pads every 10k to 15k miles or
similar, found in the vehicle’s manual. Therefore, it is expected that using a physics
based model, yet a simplified model compared to an elaborate 3D finite element model
as shown in [7], a more accurate prediction can be made. The goal is to predict the wear
so that at the next oil change pads can be replaced.
The architecture of the brake pad prognostics system is shown in Fig. 3. In general,
the characteristics of the wear on the pad depend on many variables. For the aim of
prognostics, a simplified model will be used, keeping in mind that machine learning
will be applied later to improve the model’s accuracy. In this simplified lumped model,
the wear on the brake pad is a function of the instantaneous temperature of the pad and
the instantaneous energy extracted (the power) from the vehicle’s kinetic energy as heat
[6]. In a lab environment, one may create a setup with the average temperature of the
pad being controlled so that it only fluctuates in a range DT (such as 50 °C fluctua-
tions), and the pad pressure is applied to a rotating disk for measurable durations, and a
table similar to Table 1 is generated. The pad mass is modelled by
dT P
¼ r ð T T1 Þ ð1Þ
dt mC
dm
¼ wðT; PÞ ð2Þ
dt
Connected Vehicle Prognostics Framework for Dynamic Systems 7
Here, w(T, P) is the wear from Table 1 and T∞ = Tamb. Initial conditions are
T = Tamb, m = m0, and mC is the mass multiplied by specific heat capacity of the brake
pad, and is assumed to be constant for a short duration of time, measured in days. It is
assumed that the processor solving Eq. (1) does not have access to w(T, P), and
therefore it cannot calculate the change in the mass instantaneously. The temperature
T in Eq. (1) is initialized to Tamb at ignition. This model’s accuracy will deteriorate
when almost all the mass of the pads is worn out, but at that time, the algorithm will
already have predicted that it is time to change the brake pads. Although this
approximation has errors, due to the fact that big data and machine learning will be
used as will be shown later, the error is expected to be compensated for, since it is not a
random error.
Table 1. Showing w(T, P). Here, ai,j (g/s) is the average grams lost given a power at a average
temperature for 1 s.
w Temperature (Celsius)
[-40 0] [0 40] [40 80] …
Power (Watts)
The power P is derived from vehicle speed and the friction brake torque for each
wheel after considering the loading effect ratios and the heat partitioning factor. Based
on the calculated temperature and power at a given time t, for given (T, P), Table 1,
provides the wear factor w(T,P), so that the instantaneous wear is:
Here, W is the total wear, in grams, of the brake pad until a final time tf.. Both
Eqs. (3) and (4) which is dependent on material properties will be solved in the cloud.
The wear in this model is a function of two variables, the state T and the input P. To
solve the model, create a 2D mesh in the vehicle’s memory with T and P as axes,
similar to Table 1. Set a minimum and maximum for each axis:
Partition each axis into non-intersecting intervals. Smaller partitions give higher
accuracy but require more storage. Each axis may have different number of partitions.
8 O. Makke and O. Gusikhin
This will create a 2D mesh with cells in it. Let set EM be the set of all points M which
lie in the same cell. Suppose that the algorithm runs at a rate s ms. (e.g. 5 ms). Then, at
every s ms, calculate using Eq. (1) both P(t), and T(t) where t is the current time.
Locate the set EM in the 2D mesh such that EM contains point M with coordinates (P(t),
T(t)). Then increment:
In order to utilize big data and machine learning, the data in the 2D mesh is sent to
the cloud. The data has a known size in terms of kilobytes, and hence the requirements
can be known before implementing it in the vehicle. The final architecture of the
dynamic system is shown in Fig. 5.
Fig. 4. Clustering a hypothetical curve based on both T and P. The intensity of each cell in the
2D mesh represents the sum of PðtÞs at different values of t, at coordinate (P,T).
The cloud can use big data and machine learning to derive the actual coefficients w
(T,P) if field measurements can be provided. To achieve this, experimental vehicles can
be equipped with sensors which generate data for the cloud to learn the wear factors for
each cell in Fig. 4. Alternatively, measurements can be taken during a service visit at
the dealer. Then Table 1 is constructed using machine learning and is made available in
the cloud.
Connected Vehicle Prognostics Framework for Dynamic Systems 9
Fig. 5. Brake prognostics algorithm is separated between vehicle and cloud. Here, m is assumed
constant for short durations in the vehicle.
This section addresses prognostics of components which can be represented using non-
linear models. It is desired to have a universal approach for these components. In this
approach, we would like to reduce the number of experiments conducted in the lab to
characterize the wear of the components, and replace them by using connected vehicles
and machine learning. Without loss of generality, the initial state of the system is
assumed to be known for a new part, and is 0. Suppose that the state space model for a
system is
Under the assumption that: if the degradation of the states which are used in
prognostics is slow compared to the dynamics of the system, then the states repre-
senting the degradation are assumed to be constant for that duration (a day or a week),
then the input vector x* can be separated into:
2 3
x
x ¼ ½x0 ; x1 ; . . .; xn jbx 0 ; bx 1 ; . . .; bx m T ¼ 4 5 ð10Þ
b
x
@ xbk
where 1 for 0 k m ð11Þ
@t
And then all xbk can be treated as constants for short periods of time.
W_ in Eq. (9) represents the instantaneous wear, and hence W is the total wear. The
goal is to calculate W on a different processor (in the cloud) than that which calculates
x*.
W is a vector of total wear of several subcomponents of the component under
monitor, and intuitively, we can select the worst value to be an answer. The correct
answer will depend on the nature of the problem and its impact on customers.
The following qualitative conclusion can then be made: The state Eq. (8) captures
how the component is being used, and the output Eq. (9) calculates a measure of
interest, based on the usage. The vehicle calculates and aggregates the behavioral
model, while the cloud calculates any y of interest by using a corresponding g(x*,u),
which itself can be learnt using big data and a true measurement of y using either a
sensor or manual measurement of selected few components. Algorithm 1 is used to
aggregate the data in the vehicle.
The equations are discretized and the errors are assumed to be negligible since a
customer has few weeks to change the brake pads, oil, air filter, and not few minutes. In
Connected Vehicle Prognostics Framework for Dynamic Systems 11
order to calculate W from Eq. 9, instead of integrating g(x*,u) over time, by summing
the values in time, sample by sample, we partitioned the range of g(x*,u) into cells
where each cell represents accumulation of values of W _ at same pair (x*, u), and
summed the values in the partition (the is commutative). This is described in details as
follows.
By integrating Eq. (9), W can be obtained by summing over all samples N as
X
N
W¼ gðxðti Þ; uðti ÞÞs ð12Þ
i
Since the summation is commutative, collect the values of g(x,u) at different times
together for which the pair (x,u) are similar and call this set Si. Therefore, Si is a
collection of values ci. Define gSi to be the sum of the values in set Si.
gSi ¼ ni ci ð14Þ
But ni has already been calculated in the vehicle, in step 11 of the algorithm, as EM
and therefore, the wear W is found to be
X
W¼ ci Ei s ð16Þ
i
Note that the slow changing states, which are used for prognostics, are assumed to
be constants over short durations. But, whenever the cloud updates the wear prediction,
it calculates new values for xbk and sends them to the vehicle. Hence the model
approximation remains dynamic, running at two different rates, yielding a better
approximation than assuming that these states are always constant. The wear calcu-
lation then feeds into a time prediction algorithm which completes the prognostics for
the component. The time prediction algorithm can vary in complexity as needed, and
this not an issue because it runs in the cloud where resources are not as constraint as the
vehicle. Figure 6 shows the machine learning system in the cloud.
The last remaining part is to derive the values of ci in the cloud. By either providing
manual wear measurements W, or inserting sensors in a subset of the vehicles to
12 O. Makke and O. Gusikhin
measure this wear, we can use machine learning and optimization to derive the outputs
ci. Moreover, Eq. (16) has the form of a linear regression which also resembles a
simple neural network where the activation function is just a pass-through node and in
this case, the weights of the network are the coefficient ci. and the inputs are Eis.
Fig. 7. The proposed prognostics system. The algorithms are executed on two compute entities,
in the vehicle and in the cloud.
If needed, extra inputs can be added to the system in Fig. 6, such as geographical
location, season, vehicle type, etc. when a neural network is used to add more infor-
mation to the system to reduce errors unaccounted for in the models, such as effects of
salt, dust, etc.
In this section, we will illustrate the application of the framework to cabin air filter
prognostics. Given air velocity Uin of the air feeding into the air filter, and concen-
tration of particulates before the air filter Nin, it can be shown [9] that the rate of change
of accumulation of particles, A, in the air filter to be:
Connected Vehicle Prognostics Framework for Dynamic Systems 13
dA
¼ Uin Nin ð1 Pð AÞÞ ð17Þ
dt
In Eq. (17), P(A) is the filter penetration defined as the ratio of the particle con-
centration after the air filter over the particle concentration before the inlet.
Nout
Pð AÞ ¼ ð18Þ
Nin
The wear here is A, the accumulation of particles in the air filter, and the larger A,
the more wear. By integrating Eq. (17) and assuming initial wear is 0, the total wear
can be found to be:
Zt
AðtÞ ¼ Uin ðsÞNin ðsÞ½1 PðAðsÞÞds ð19Þ
0
Nout
Nin ¼ ð20Þ
Pð AÞ
The system’s inputs are blower speed B, recirculation door position R, the con-
centration Nin, and the vehicle speed V, and the states are air velocity into the air filter
Uin, the concentration after the filter Nout, and the wear A. A is assumed to be slow
changing for the duration of few hours or days, depending on the region. Moreover,
here, Uin is assumed to be approximated as a function with arguments A, V, B and R,
and that it is not a dynamic system. This assumption is equivalent to saying that the
steady state response of Uin after changing any of the inputs is reached quickly. Under
this assumption, only A, V, B and R are needed to be known. Therefore,
known by the OEM. The OEM also has derived the mapping between A and P
(A) based on experimental data. The cloud calculates the wear for each cell Ei of value
ni in the hyperrectangle
And then, the total wear A is found by summing the individual wears in each cell,
which happens in the cloud as shown in Eq. (24). The parameters except ni are known
from the axes coordinates, and ni is accumulated in the vehicle.
X
Cells
Nout
A¼ f ðA; B; R; V Þ ð1 Pð AÞÞni ð24Þ
i
Pð AÞ
After executing Eq. (24), a new value of P(A) is obtained from the already existing
lookup table. The problem reduces to knowing the function f. This function can be
obtained by using neural networks as global approximators, and data can be generated
before production by measuring in a vehicle the air velocity under different conditions.
The system is shown in Fig. 8. The function f is chosen to be in the cloud since it can
be improved over time. The values of A and P(A) are sent to the vehicle for use in other
features.
6 Conclusion
data analysis and dynamic systems theory to create a hybrid solution. Furthermore, the
theoretical insight is provided to show what needs to be aggregated in the vehicle, what
assumptions are made, and how to map this data to machine learning inputs. The data
generation is simplified to aggregating the time in a hyperrectangle which has the states
and inputs of the dynamic system as axes. The application of the framework to two
dynamic models: brake pad wear prediction and in cabin air filter life, is discussed. If
more accuracy is desired, more elaborate models can be applied under same
framework.
The cloud system can use machine learning to predict the time to service com-
ponent, which can combine both machine learning and input from customers’ calendar
information to optimize the service scheduling for the vehicle.
This approach provides resistance to connectivity issues since the aggregated data
in the vehicle has a long time window, in terms of days, to connect once to the cloud to
transfer the aggregated data and obtain a new wear value.
The verification of the accuracy of the models still requires large set of field data.
Upon receiving sufficient amount of data, we will analyze the prediction and if nec-
essary, we will apply higher fidelity models. The future work includes transforming
other models of components and subsystems such as fuel pumps, tires, and batteries to
this framework.
References
1. Gusikhin, O., Rychtyckyj, N., Filev, D.: Intelligent systems in the automotive industry:
applications and trends. Knowl. Inf. Syst. 12, 147–168 (2007)
2. Siegel, J., Erb, D., Sarma, S.: A survey of the connected vehicle landscape–architectures,
enabling technologies, applications, and development areas. IEEE Trans. Intell. Transp. Syst.
19, 2391–2406 (2017)
3. Zhang, Y., Gantt Jr., G., Rychlinski, M., Edwards, R., Correia, J., Wolf, C.: Connected
vehicle diagnostics and prognostics, concept and initial practice. IEEE: Trans. Reliab. 58,
286–294 (2009)
4. Zagajac, J., Chopra, A., Krivtsov, V., Gusikhin, O.: Method and apparatus for connected
vehicle system wear estimation and maintenance scheduling. US Patent Application
US2016/0163130 A1 (2016)
5. Pecht, M.: Prognostics and Health Management of Electronics. Wiley-Interscience, New York
(2008)
6. Zhang, S., Chen, W., Li, Y.: Wear of friction material during vehicle braking. In: SAE
Technical Paper 2009-01-1032 (2009)
7. Bakar, A.R.A., Ouyang, H., Khai, L.C., Abdullah, M.S.: Thermal analysis of a disc brake
model considering a real brake pad surface and wear. Int. J. Veh. Struct. Syst. 2(1), 20–27
(2010)
8. Kambhampati, C., Graces, F., Warwick, K.: Approximation of non-autonomous dynamic
systems by continuous time recurrent neural networks. In: Proceedings of the IEEE-INNS-
ENNS International Joint Conference on Neural Networks (2000)
9. Billings, E.C.: Effects of particle accumulation in aerosol filtration. Ph.D. dissertation,
California Institute of Technology. http://resolver.caltech.edu/CaltechETD:etd-09172002-
113217. Accessed 14 May 2018
Another random document with
no related content on Scribd:
manufacture of textile machinery he added that of general machinery
and large tools for cutting, boring, rifling, planing and slotting. He had
a great reputation in his day, but his work seems to have been more
that of a builder of standard tools than an originator of new tools and
methods.
Charles Holtzapffel, another well-known engineer of that
generation, was the son of a German mechanic who came to
London in 1787. He received a good education, theoretical as well
as practical, and became a skilled mechanician and a tool builder of
wide influence. His principal book, “Turning and Mechanical
Manipulation,” published in 1843 in three volumes, is an admirable
piece of work. Covering a field much wider than its title indicates, it is
the fullest and best statement of the art at that time; and scattered
through it there is a large amount of very reliable mechanical history.
By 1840 the number of men engaged in tool building was
increasing rapidly, and it is impossible to consider many English tool
builders who were well known and who did valuable work, such as
Lewis of Manchester, B. Hick & Son of Bolton, and others. One
noteworthy man, however, ought to be mentioned—John George
Bodmer, who was neither an Englishman, nor, primarily, a tool
builder.[71] He was a Swiss who worked in Baden and Austria, as
well as in England, and his fertile ingenuity covered so many fields
that a list of the subjects covered by his patents occupy six pages in
the “Transactions of the Institution of Civil Engineers.”
[71] For a “Memoir” of Bodmer see “Transactions of the Institution of Civil
Engineers,” Vol. XXVIII, p. 573. London, 1868.
Bodmer went to England for the first time in 1816 and visited all
the principal machine shops, textile mills and iron works. He returned
in 1824 and again in 1833, this time remaining many years. On his
second trip he established a small factory for the manufacture of
textile machinery at Bolton, in which was one of the first, if not the
first, traveling crane.[75] At the beginning of his last and long
residence in England, Bodmer appointed Sharp, Roberts &
Company makers of his improved cotton machinery, which they also
undertook to recommend and introduce. This arrangement was not
successful, and a few years later, in partnership with Mr. H. H. Birley,
Bodmer started a machine shop and foundry in Manchester for
building machinery.
[75] Ibid., p. 581.
Nearly all of the machinery for the Manchester plant was designed
and built by Bodmer himself and it forms the subject of two
remarkable patents, granted, one in 1839 and the other in 1841.[76]
The two patents cover in reality nearly forty distinct inventions in
machinery and tools “for cutting, planing, turning, drilling, and rolling
metal,” and “screwing stocks, taps and dies, and certain other tools.”
“Gradually, nearly the whole of these tools were actually constructed
and set to work. The small lathes, the large lathes, and the planing,
drilling, and slotting machines were systematically arranged in rows,
according to a carefully-prepared plan; the large lathes being
provided, overhead, with small traveling cranes, fitted with pulley-
blocks, for the purpose of enabling the workmen more economically
and conveniently to set the articles to be operated upon in the lathes,
and to remove them after being finished. Small cranes were also
erected in sufficient numbers within easy reach of the planing
machines, &c., besides which several lines of rails traversed the
shop from end to end for the easy conveyance on trucks of the parts
of machinery to be operated upon.”[77] There were, in addition to
these, however, “a large radial boring machine and a wheel-cutting
machine capable of taking in wheels of 15 feet in diameter, and of
splendid workmanship, especially in regard to the dividing wheel,
and a number of useful break or gap-lathes, were also constructed
and used with advantage. It is especially necessary to mention a
number of small, 6-inch, screwing lathes, which, by means of a
treadle acting upon the driving gear overhead, and a double slide-
rest—one of the tools moving into cut as the other was withdrawn,—
screw cutting could uninterruptedly proceed both in the forward and
in the backward motion of the toolslide, and therefore a given
amount of work accomplished in half the time which it would occupy
by the use of the ordinary means. Some of the slide-lathes were also
arranged for taking simultaneously a roughing and finishing cut.”[78]
[76] The first of these is described in the American Machinist of March
13, 1902, p. 369.
[77] “Memoir,” p. 588.
[78] Ibid., p. 597-598.
The latter part of Bodmer’s life was spent in and near Vienna,
working on engines and boilers, beet sugar machinery and
ordnance; and at Zurich, where he died in 1864, in his seventy-ninth
year.
Bodmer does not seem to have originated any new types of
machine tools, with the exception of the vertical boring-mill, which he
clearly describes, terming it a “circular planer.” It was little used in
England, and has been considered an American development.
It is hard now to determine how far Bodmer has influenced tool
design. It was much, anyway. Speaking of the patent just referred to,
John Richards, who has himself done so much for tool design, says,
“Here was the beginning of the practice that endured.” He has
described some of Bodmer’s tools in a series of articles which show
a standard of design greatly in advance of the practice of his time.[79]
Another writer says of Bodmer, “He seems always to have
thoroughly understood the problems he undertook to solve.” “One is
lost in admiration at the versatility of the inventive genius which could
at any one time—and that so early in the history of machine design
—evolve such excellent conceptions of what was needed in so many
branches of the mechanics’ art.”[80]
[79] American Machinist, Vol. XXII, pp. 352, 379, 402, 430, 457, 478,
507, 531, 559, 586, 607, 637.
[80] Ibid., Vol. XXV, p. 369.