Content uploaded by Angelo Leogrande
Author content
All content in this area was uploaded by Angelo Leogrande on Jan 31, 2022
Content may be subject to copyright.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
DOI: 10.5121/ijaia.2022.13103 37
PREDICTIVE MAINTENANCE AND ENGINEERED
PROCESSES IN MECHATRONIC INDUSTRY: AN
ITALIAN CASE STUDY
Alessandro Massaro1,2,*, Gabriele Cosoli1,
Angelo Leogrande1 and Nicola Magaletti1
1LUM Enterprise srl, S.S. 100 - Km.18, Parco il Baricentro, 70010, Bari, Italy
2LUM - Libera Università Mediterranea "Giuseppe Degennaro",
S.S. 100 - Km.18, Parco il Baricentro, 70010, Bari, Italy
ABSTRACT
The paper proposes the results of a research industry project concerning predictive maintenance process
optimization, applied to a machine cutting polyurethane. A company producing cutting machines, has been
provided with an online control system able to detect blade status of a machine supplied to a customer
producing polyurethane components. A software platform has been developed for the real time monitoring
of the blade status and for the prediction of the break up conditions adopting a multi-parametric data
analysis approach, based on the simultaneous use of unsupervised and supervised machine learning
algorithms. Specifically, the proposed method adopts a k-Means algorithm to classify bidimensional risk
maps, and a Long Short Term Memory (LSTM) one to predict the alerting levels based on the analysis of
the last values for some process variables. The analysed algorithms are applied to an experimental dataset.
KEYWORDS
Decision Support System, Process Engineering, Sales Prediction, Artificial Intelligence.
1. INTRODUCTION
The technology innovation of mechatronic systems combined with Artificial Intelligence (AI)
facilities is an important research topic in industrial engineering [1]. The case study of the
proposed paper is addressed on this main topic, focusing the attention on predictive maintenance
tools which can be performed mainly to avoid machine breakdowns and product defects [2]-[4].
In this scenario, different sensors can be adopted to detect cutting machine data for monitoring
blade status. Concerning manufacturing processes, the approach to monitor wear status can be
based on acoustic multi-sensors systems [5] as well as on Artificial Neural Networks (ANNs)
able to estimate and classify certain wear parameters [6]. Cutting tool wear analysis can be
performed also by microscope-based 3D image process too, providing the blade wear profile [7].
Some studies highlight that wear conditions can be analysed by the relationship between
temperature and electrical resistance [8], or defining wear classes applying thermography
combined to Convolutional Neural Network (CNN) [9]. In particular, AI Elman Adaboost
approaches are used to predict wear conditions, by analysing force data, vibration data, acoustic
emission signal, and other multi-sensor data [10]. Cutting forces and vibrations are surely
important parameters to detect wear [11]. Temperature distribution analysis [12] can be useful to
understand physical phenomena such as elongation in metallic components [13],[14]. Machine
learning unsupervised and supervised algorithms, such as respectively k-Means [15] and Long
Short Term Memory (LSTM) [16], are suitable for predictive maintenance applications, thus
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
38
suggesting their use for this specific case study. All the variables can be processed
simultaneously to find criteria oriented on predictive maintenance of the whole cutting machines,
and of each part such as the blade component. With the aim of undertaking an innovative
business model based on customer care, industries producing cutting machines could provide
predictive maintenance services by real time monitoring and AI data processing. The pilot
company, FEMA. srl, is addressed on these services suitable to predict and reduce failures of the
machines cutting polyurethane. At the beginning of the company activity, the maintenance or
replacement of a component was activated only after a failure occurred and often when the
component has reached the end of its life cycle. An unexpected machine downtime, seriously
affects the progress of the production process, resulting in expensive consequences such as: (i)
decrease of the Overall Equipment Effectiveness (OEE) of the machine and / or plant; (ii)
damages (eg higher expenses for overtime work, lower revenues); (iii) delays in the production
plan and in the fulfilment of orders; (iv) long production stops, if there is no availability in the
warehouse of the spare parts necessary for the immediate repair of the machinery; (v) end
customer dissatisfaction. To avoid these risks, the pilot company producing cutting machines is
oriented to provide an advanced predictive maintenance service adopting some of the results
achieved with the Smart District 4.0 (SD 4.0) project. SD 4.0 is a project supported by the Italian
Ministry of Economic Development (MISE), with the aim of stimulating the widespread
digitization processes of Small and Medium-sized Enterprises (SMEs) in some typical sectors as
mechatronic. The project provides as "deliverables", different technologies discussed in this work
including software platform interface, data warehouse system, and application of machine
learning algorithms. Specifically, the paper is structured in the following steps:
• definition of the main architecture of the pilot application describing the cutting machine to
control in cloud;
• AS-IS and TO-BE process mapping, by highlighting how technologies improve the
predictive maintenance services;
• design of the data flow Unified Modeling Language (UML) diagram, of the SD 4.0
platform, describing all the functions of the actors involved in the TO-BE process;
• discussion of a multi-parametric analysis by unsupervised k-Means algorithm providing bi-
dimensional risk maps based on the simultaneous analysis of "key-variables" such as blade
temperature, stretch and speed;
• prediction by LSTM approach, of the blade status analysing the last sensor data and
allocation of the predicted clustered results into the risk maps.
2. ARCHITECTURE DESIGN AND BPMN PROCESSES
In this pilot application of the SD 4.0 project, it was decided to decline the use of the project IT
platform in the context of predictive maintenance of cutting machines, proposing changes to the
current AS-IS working methods often adopted by such companies (i.e.: sofa and other padded
products manufacturers). The goals of the implementation of the new business model are to use
the platform as a collaboration tool for the entire supply chain and, through the application of
predictive maintenance, to predict failures, plan maintenance, reduce downtime and maximize
OEE. The TO-BE operating model for the predictive maintenance is mainly sketched in Fig. 1
indicating the system’s actors: a customer company purchases the cutting machine from the pilot
company, by connecting this machine to the SD 4.0 cloud platform to acquire in real time the
data streaming useful for blade failure prediction. The platform sends to the customer company
forecasts about the need to carry out maintenance. In the event that there is a need for
extraordinary maintenance or spare parts, the customer company, through the platform, can notify
the pilot company, which can thus combine the sale of the machine, with the sale of the
maintenance and predictive maintenance services, to be provided through the platform. The
customer company can innovate the production and make its process more efficient by getting
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
39
some benefits such as receiving alerts about machine status and planning timely interventions,
with a significant reduction in costs because of failure prediction. Figure 1 shows the data flow of
the connected machine, where the supplier of machines can run some analysis, obtaining useful
information to increase knowledge about the cutting machine in different operating conditions
with the aim of achieving product improvement as well as getting relevant inputs for the redesign
of the whole machine or of some of its components. The pilot company also has the possibility of
optimizing the management of warehouse stocks for consumables and spare parts, having the
ability to predict how many blades are reaching the end of their life and, consequently, to avoid
stockout conditions.
Figure 1. Main architecture of the SD 4.0 platform integrating new TO BE business process oriented on
predictive maintenance.
In the Business Process Modeling Notation (BPMN) model of Fig. 2, are illustrated the
interactions between AS-IS supply chain actors highlighting the performed activities, and the
tools used for communications. To date, the fabricated machines are not connected to the
production company factory, and the telephone channel is mainly used for communications
between the pilot company and customers. The Computerized Numerical Control (CNC)
machines built by FEMA srl company, are equipped by different sensors and actuators, by
allowing the timely detection of a fault, in order to prevent a significant damage. The sensors
currently installed on most machines are with Internet of Things (IoT) connectivity, by adopting
specific data protocols [1]. The experimental CNC machine is the Giotto EVO [17], which is
connected to a cloud platform. The Giotto EVO CNC machine is an electronic shaper for cutting
foam resins, rigid, semi-rigid and flexible polyurethane foam. Available in three versions, it is
equipped with two cutting systems with an oscillating blade and a rotating blade. These two
cutting systems offer a high level of cut quality and maximum speed on all densities and textures,
from the lowest and softest to the highest and most rigid. The machine integrates the following
sensors:
a blade presence sensor that is able to detect if the blade is out of place: if the blade breaks
or comes out of one of the flywheels, the sensor sends an alarm signal to the PLC which
blocks the machine;
a blade tensioning device that allows to keep the tension on the blade within a certain range
(very important aspect to perform a clean and precise cut);
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
40
a blade cooling nozzle that allows to control the temperature during the processing phases
(very important to avoid the overheating of the blade);
a video control system of the blade sharpening which, by means of a double camera,
allows to detect, with the machine stopped, the need to carry out the automatic sharpening
of the blade.
Currently, a “reactive” maintenance policy is carried out only when a fault occurs. In particular,
in this process, the failure taken into consideration is the breaking of the blade. The process
begins in the customer company's Pool of Fig. 2, when a machine failure occurs during the
production activity. When a blade breaks, the company immediately stops production and, having
verified the type of fault, evaluates if it is possible to proceed with the blade change, ordinary
maintenance, or if it is necessary to request the intervention for an extraordinary maintenance
intervention or to request a spare part. The FEMA srl company, having received and analysed the
request from the customer company, provides the service and sends the necessary spare part if
available in stock. Once the maintenance interventions are terminated, the customer company can
resume production. Thanks to the predictive algorithms, the SD 4.0 platform can estimate the end
of life of each blade so as to be able to schedule its replacement in time, making the most of its
useful life, before the failure occurs. In this way it would be possible to carry out maintenance
when production is stopped. In the AS-IS process, each time a blade breaks, the operator checks
that it has not damaged other parts of the machine. In the event of chain effects, an intervention
by the manufacturer of the machine and possibly spare parts may be required, greatly extending
the downtime. In the AS-IS process, at each break of a blade, the operator must discard the
polyurethane block on which the machine was performing the cut. The rejection of the
polyurethane block constitutes a further economic damage for the customer company. In the AS-
IS process, all requests for technical assistance from customers are urgent as they arise from a
failure. All requests for assistance arrive to the supplier via the telephone channel and it may
happen that operators are busy carrying out other activities, thus having to wait for the
intervention, and extending the downtime. The platform enables the continuous monitoring of the
machine by generating automatic alerts in the event of anomalies, completely replacing the
telephone channel, and guaranteeing timely interventions. To plan the maintenance intervention,
the pilot company must check the availability of the operators and / or the necessary spare parts.
If the skilled workers are employed in other activities, or spare parts are not available in the
warehouse, the process is blocked by introducing delays in the production plan of the client
company. These problems are very serious as the failure has already occurred, production is
stopped and delays, in the most serious cases, can lead to the loss of entire production days.
Figure 2. BPMN AS-IS process schematization.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
41
To achieve the improvement goals of the AS-IS process, it is necessary to redesign the
maintenance process, moving from a “reactive” policy to a predictive policy (TO-BE). The main
actors involved in the TO BE process (see Fig. 3) are: the FEMA srl company providing
predictive maintenance services, workers to perform services, the customer company (Polirex
srl), and the SD 4.0 platform.
The analysis carried out on the predictive maintenance process to be implemented, focuses on the
ability to predict and reduce failures through maintenance. The process begins in the Pool of the
customer company with the start of production (beginning of the polyurethane cutting
operations). During each process, a check is performed cyclically on the status of the production
in progress, providing the following outcome:
production successfully completed (the cutting operation of the polyurethane block are
completed without any problems);
production in progress without any errors or failure (the cutting operation is still in
progress and no problem or failure has occurred.
In these cases, the machine used in the customer company, cyclically acquires the data and
automatically sends them to the SD 4.0 platform, through a module able to data importing and
data processing. The SD 4.0 platform (Service Provider Pool) collects the data useful also to train
the algorithms predictive models, and returns the real time control and the machine diagnosis.
The customer company performs a check on the diagnosis just received by verifying the forecast
of failure, and, if a failure is expected, the company verifies the need to make or not an order
request for one or more spare parts. If the spare parts are not necessary, the company plans the
maintenance intervention, otherwise it proceeds with sending the order request to the machine
supplier. The FEMA srl company verifies the availability of spare parts and/or personnel required
for maintenance, and then plans in detail the intervention (times, costs, etc.). The customer
company receives the details of the supply and consequently schedules the predictive
maintenance intervention. The cutting process can be interrupted in the following two cases:
production interrupted by an unexpected break (the machine sends data to the service
provider activating an alerting);
production to be interrupted for a planned maintenance intervention.
In both the two cases, once maintenance has been performed, production and the entire process
can be restarted. The recording and classification of the process data detected in these
circumstances enriches the experimental dataset improving the machine learning models. In Fig.
3 is illustrated the whole BPMN TO BE described process.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
42
Figure 3. BPMN “TO BE” process involving SD 4.0 platform, customer, and Fema s.r.l. company.
The detected data enrich the experimental dataset by improving the machine learning models. In
Fig. 3 is illustrated the whole BPMN TO BE described process, involving SD 4.0 platform,
customer, and Fema s.r.l. company. The Unified Modeling Language (UML) scheme of Fig. 4
indicates the Use Case Diagram (UCD), indicates all system functions and actors, and shows SD
4.0 platform data flow. In the diagram is distinguished the data warehouse system (Google Cloud
BigQuery), and the relationships between all actors (FEMA srl, worker indicated for the
maintenance, monitored machine, and customer) acting in the TO BE process sketched in Fig. 3.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
43
Figure 4. UML UCD diagram of the SD 4.0 platform data flow.
3. SD 4.0 FRONTEND INTERFACES: REAL TIME MONITORING OF BLADE
STATUS
The SD 4.0 platform is developed by embedding different graphical dashboards monitoring blade
characteristics. In Fig. 5 is illustrated a screenshot of the main SD 4.0 interface. Each blade can
be plotted in real time with different values (see Fig. 6), such as size and weight of the piece to
cut, the blade temperature, the blade strength, and the blade speed. The platform also allows the
plotting of the measured parameters as histograms indicating the variables distribution (see Fig.
7). The interface provides further blade information such as average values and the overcoming
of threshold conditions as a multi-parametric alerting system. As the primary choice of the
alerting condition, is the real time check of the overcoming threshold condition. The interface
allows the data monitoring and filtering, by means of the selection of the dataset of different
blades. The sensor sampling time is about 1 second. Every hour all data collected into packets,
are transmitted to the platform backend updating the experimental dataset. The estimated average
life time of a blade is ranging from 11 to 38 days.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
44
Figure 5. Dashboard filtering data of the different blades adopted for the manufacturing process.
Figure 6. Platform dashboard: time domain plotting of Kg, Speed, temperature, stretch and size parameters.
Figure 7. Examples of variable dashboards plotted on the same interface: KG, stretch, temperature and size
distributions.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
45
4. DATA PROCESSING OF THE EXPERIMENTAL DATASET
The experimental dataset is related to seven blades blade (from April to July 2021 concerning the
substitution of more blades), and contains the following attributes for total number of 710.000
records for the specific analysed blade (from 1 June 2021 to 9 July):
Max Life (maximum life time of the blade in terms of observations before the occurrence
of the event that determines its breakage or replacement);
Average Life (average life time of the blade);
Average Kg (average weight of the polyurethane pieces that are placed on the machine in
order to be cut);
Average Stretch (average blade stretch expressed in mm);
Average Temperature (average blade temperature expressed in °C);
Average Size (average size expressed in cm);
Average Speed (average blade frequency in Hz units);
Average Sharpening (Boolean status; if the sharpener is on, the value assigned is 1,
otherwise it is zero).
In Fig. 8 is illustrated an example of the experimental dataset concerning the stretch of a
monitored blade, where the low values indicate standby conditions (phases between two cutting
processes named cycles). As observed by the stretch trend of Fig. 8, the stretch during the time
increases: this indicates an irreversible elongation of the blade.
Figure 8. Example of data, extracted from the whole experimental dataset concerning stretching of a single
blade (710.000 records).
The first approach to follow is to define the risk maps by considering couples of key-parameters
such as blade stretch, blade temperature and blade speed. The Konstanz Information Miner
(KNIME) [1] workflow of Fig. 9, is able to provide three clusters of experimental data, by
applying the k-Means algorithm. The three clusters indicate three conditions: no alerting
condition, weak alerting condition, and the strong alerting condition.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
46
Figure 9. KNIME k-Means workflow used for the definition of the risk map layouts.
In Fig. 10 is illustrated the clustering results of three data clusters (cluster 0, cluster 1, cluster 3),
which are included in the box representing the three alerting conditions: the green box is the no
alerting condition, the yellow box is the weak alerting condition, and the red box is the strong
alerting conditions. The boxes represent the risk map layout and are deducted by the position of
the blue lines indicating the average measured values of the analysed variables. By plotting the
three clusters related to the temperature and stretch variables, the risk map of Fig. 10, shows the
cluster 0 as the most dangerous couple of values (strong dangerous condition). The same
“alerting” cluster 0 is identified in Fig. 11 and Fig. 12, by considering the couple of variables
speed-stretch and temperature-speed (see red quadrants), respectively. The identification of the
three alerting levels provided by the clustering algorithm, is adopted to define the risk map
layouts represented by the three different quadrants (green, yellow and red), which will be used in
order to check where the predicted couple of variables will belong. The prediction is performed
by implementing the LSTM algorithm adopting KNIME and Keras libraries by means of the
workflow of Fig. 13. By considering the last ten measured values, are predicted the couples of
variables of Fig. 14, Fig. 15, and Fig. 16 matching with the risk map layouts. Having considered
the last values before replacing the blade (having reached the condition of exceeding the
threshold values of the average values), it is observed that the values are placed in the quadrant of
greatest danger.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
47
Figure 10. k-Means results indicating clusters in the temperature-stretch plane (k=3).
Figure 11. k-Means results indicating clusters in the speed-stretch plane (k=3).
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
48
Figure 12. k-Means results indicating clusters in the temperature-speed plane (k=3).
Figure 13. KNIME workflows implementing LSTM algorithm and predicting parameters.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
49
Figure 14. Prediction of combination of temperature and stretch parameters and matching with the risk map
layout.
Figure 15. Prediction of combination of speed and stretch parameters and matching with the risk map
layout.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
50
Figure 16. Prediction of combination of temperature and speed parameters and matching with the risk map
layout.
As proved by Fig. 17, the prediction results are estimated after an enough batches number,
providing a good LSTM accuracy trend. The oscillatory trend of the accuracy parameter follows
the oscillation of the training dataset (standby conditions of the machine downtime conditions).
In Fig. 18 is sketched the LSTM network adopted for risk prediction (Keras Tensorflow libraries
implemented in KNIME objects). The used hyperparameters of the LSTM model are: input shape
=6, 50 epochs, batch size =1, 1 dense LSTM layer with 6 neurons implementing a ReLU
activation function [18],1 dense LSTM layer with 6 neurons implementing a Softmax activation
function, an output layer made by 6 neurons, RMSProp as optimizer.
Figure 17. Training accuracy versus batches.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
51
Figure 18. Model of the LSTM network predicting risks.
The proposed approach is useful also to optimize the risk maps layouts and to re-plan the
maintenance schedule of each cutting machine. The standard predictive maintenance plan could
change based on the LSTM prediction of the monitored blade variables: the planned period to
perform maintenance can be anticipated by observing a different slope behavior of the
maintenance graphs plotting the variables versus the time [19]-[20], [1]. The clustering approach
can provide interfaces [21] and further information about threshold tuning, by allowing the
optimization of the risk map layouts and updating the maintenance plan. The LSTM approach is
suitable for prediction and classification approachs in industrial applications [22],[23]. On the
other side, clustering technique can be adopted in advenced robotics [24]. Following the
industrial application scenario, k-Measn and LSTM algorithms are good candidate for Industry
5.0 environemnt, where predictive maintenace plays an importat role.
5. CONCLUSIONS
The paper proposes the results of a case study of predictive maintenance application of a research
industry project. The project is addressed on the study of a method suitable for the design of a
cloud software platform, integrating supply chain, predictive maintenance processes and
graphical alerting dashboards alerting blade wear. The alerting maps are structured by k-Means
clustering results, identifying risk layouts in the bidimensional plane. The LSTM prediction of
some couples of parameters are allocated into these risk maps, thus estimating the next wear
conditions of the blade. The proposed approach is based on the use of both unsupervised and
supervised machine learning algorithms and is applied to the specific case study testing an
experimental dataset developed within the framework of a project partially funded by the
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
52
Ministry of the Economic Development (MISE). The analysed method can be adapted to other
Industry 4.0 manufacturing processes enabling predictive maintenance processes.
ACKNOWLEDGEMENTS
All the applications have been deployed by a unique IT collaborative framework developed
within the Smart District 4.0 Project: the Italian Fondo per la Crescita Sostenibile, Bando
“Agenda Digitale”, D.M. Oct. 15th, 2014, funded by “Ministero dello Sviluppo Economico”.
This is an initiative funded with the contribution of the Italian Ministry of Economic
Development aiming to sustain the digitization process of the Italian SMEs. Authors thank to the
partner Noovle for the collaboration provided during the work development, and to Dr. Nicole
Accettura for the work developed in her thesis concerning predictive maintenance applied to the
pilot case study: “Il Machine Learning impiegato nei diversi contesti aziendali – caso di studio
sulla Predictive Maintenance”. The proposed results are used to verify the usability of the data on
the platform.
REFERENCES
[1] Massaro, A. (2021). Electronic in advanced research industry: from Industry 4.0 to Industry 5.0
advances. Wiley/IEEE book, ISBN: 9781119716877.
https://books.google.it/books?id=LP5FEAAAQBAJ
[2] Massaro, A., Maritati, V., Galiano, A., Birardi, V. & Pellicani, L. (2018). ESB platform integrating
KNIME data mining tool oriented on Industry 4.0 based on artificial neural network predictive
maintenance. International Journal of Artificial Intelligence and Applications (IJAIA), 9(3), 1-17.
http://dx.doi.org/10.5121/ijaia.2018.9301
[3] Massaro, A., Manfredonia, I., Galiano, A., Pellicani, L. & Birardi, V. (2019). Sensing and quality
monitoring facilities designed for pasta industry including traceability, image vision and predictive
maintenance. In IEEE International Workshop on Metrology for Industry 4.0 and IoT, IEEE
Proceeding (pp. 68-72). https://doi.org/10.1109/METROI4.2019.8792912
[4] Massaro, A. & Galiano, A. (2020). Re-engineering process in a food factory: an overview of
technologies and approaches for the design of pasta production processes. Production &
Manufacturing Research, 8(1), 80-100. https://doi.org/10.1080/21693277.2020.1749180
[5] Li, D., Liu, Y. & Chen, Y. (2011) (Eds.): CCTA 2010, Part IV, IFIP AICT 347, 215–220, © IFIP
International Federation for Information Processing 2011.
[6] Saglam, H. (2011). Tool wear monitoring in bandsawing using neural networks and Taguchi’s design
of experiments. International Journal of Advanced Manufacturing Technology, 55(9), 969–982.
http://dx.doi.org/10.1007/s00170-010-3133-1
[7] Čerče, L., Pušavec, F. & Kopač, J. (2015). 3D cutting tool-wear monitoring in the process. Journal of
Mechanical Science and Technology, 29(9), 3885-3895. http://dx.doi.org/10.1007/s12206-015-0834-2
[8] Sadílek, M., Kratochvíl, J., Petrů, J., Čep, R., Zlámal, T. & Stančeková, D. (2014). Cutting tool wear
monitoring with the use of impedance layers. Tehnički vjesnik, 21(3), 639-644.
[9] Klancnik, S. (2021). Automatic identification of tool wear based on thermography and a
convolutional neural network during the turning process. Sensors, 21(5).
https://doi.org/10.3390/s21051917
[10] Liu, Y., Wang, F., Lv, J. & Wang, X. (2020). A novel method for tool identification and wear
condition assessment based on multi-sensor data. Applied Sciences, 10(8).
https://doi.org/10.3390/app10082746
[11] Xu, C., Dou, J., Chai, Y., Li, H., Shi, Z. & Jing, X. (2018). The relationships between cutting
parameters, tool wear, cutting force and vibration. Advances in Mechanical Engineering, 10(1), 1–14.
https://doi.org/10.1177/1687814017750434
[12] Massaro A. & Galiano, A. (2020). Infrared thermography for intelligent robotic systems in research
industry inspections: thermography in industry processes. IGI Global, Handbook of Research on
Advanced Mechatronic Systems and Intelligent Robotics, Ch. 5, ISSN: 2327-0411, pp. 98-125.
http://dx.doi.org/10.4018/978-1-7998-0137-5.ch005
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
53
[13] NagaBhushana Rao, V., Niranjan Kumar, I. N., Madhulata, N. & Abhijeet, A. (2014). Mechanical
analysis of 1st stage marine gas turbine blade. International Journal of Advanced Science and
Technology, 68, 57-64. http://dx.doi.org/10.14257/ijast.2014.68.06
[14] Guo, Y., Ye, W., Xu, X. (2021). Numerical and experimental investigation of the temperature rise of
cutting tools caused by the tool deflection energy. Machines, 9(6).
https://doi.org/10.3390/machines9060122
[15] Uhlmann, E., Pontes, R. P., Geisert, C. & Hohwieler, E. (2018). Cluster identification of sensor data
for predictive maintenance in a selective laser melting machine tool. Procedia Manufacturing, 24, 60-
65. https://doi.org/10.1016/j.promfg.2018.06.009.
[16] [16] Bruneo, D. & De Vita, F. (2019). On the use of LSTM networks for predictive maintenance in
smart industries. In IEEE International Conference on Smart Computing -SMARTCOMP-, (pp. 241-
248). https://doi.org/10.1109/SMARTCOMP.2019.00059
[17] Giotto Evo. https://femaindustry.com/product/giotto-evo/
[18] Massaro, A. State of the Art and Technology Innovation. In Electronics in Advanced Research
Industries: Industry 4.0 to Industry 5.0 Advances, IEEE, 2022, pp.1-49.
https://doi.org/10.1002/9781119716907.ch1
[19] Massaro, A., Selicato, S. & Galiano, A. (2020). Predictive maintenance of bus fleet by intelligent
smart electronic board implementing artificial intelligence. IoT, 1(2), 180-197.
https://doi.org/10.3390/iot1020012 .
[20] Massaro, A. Information technology infrastructures supporting Industry 5.0 facilities. In Electronics
in Advanced Research Industries: Industry 4.0 to Industry 5.0 Advances, IEEE, 2022, pp.51-101.
https://doi.org/10.1002/9781119716907.ch2
[21] Massaro, A. Human–Machine interfaces. In Electronics in Advanced Research Industries: Industry
4.0 to Industry 5.0 Advances, IEEE, 2022, pp.103-153. https://doi.org/10.1002/9781119716907.ch3
[22] Massaro, A. Internet of Things solutions in industry. In Electronics in Advanced Research Industries:
Industry 4.0 to Industry 5.0 Advances, IEEE, 2022, pp.155-202.
https://doi.org/10.1002/9781119716907.ch4
[23] Massaro, A. Image vision advances. In Electronics in Advanced Research Industries: Industry 4.0 to
Industry 5.0 Advances, IEEE, 2022, pp.301-340. https://doi.org/10.1002/9781119716907.ch7
[24] Massaro, A. Advanced robotics. In Electronics in Advanced Research Industries: Industry 4.0 to
Industry 5.0 Advances, IEEE, 2022, pp.203-252. https://doi.org/10.1002/9781119716907.ch5
AUTHORS
Alessandro Massaro. Professor Alessandro Massaro (ING/INF/01, FIS/01, FIS/03)
carried out scientific research at the Polytechnic University of Marche, at CNR, and
at Italian Institute of Technology (IIT) as Team Leader by activating laboratories for
nanocomposite sensors for industrial robotics. He is in MIUR register as scientific
expert in competitive Industrial Research and social development. He was the head
of the Research and Development section and scientific director of MIUR Research
Institute Dyrecta Lab Srl. Member of the International Scientific Committee of
Measurers IMEKO and IEEE Senior member, he received an award from the
National Council of Engineers as Best Engineer of Italy 2018 (Top Young Engineer 2018). He is currently
researcher at LUM Enterprise srl, and professor at LUM University Libera Università Mediterranea
"Giuseppe Degennaro".
Gabriele Cosoli. Senior IT Specialist and Solution architect with a degree in Computer
Science and over five years of previous experience, specialized in the analysis and
design of ICT solutions in various application areas and technological frameworks.
Certified on "Machine Learning by Stanford University on Coursera" Master on "Agile
and Digital Project Management - Adavanced Course" at 24ORE Business School.
International Journal of Artificial Intelligence and Applications (IJAIA), Vol.13, No.1, January 2022
54
Angelo Leogrande. Senior IT Specialist and Solution architect with a degree in
Computer Science and over five years of previous experience, specialized in the
analysis and design of ICT solutions in various application areas and technological
frameworks. Certified on "Machine Learning by Stanford University on Coursera"
Master on "Agile and Digital Project Management - Adavanced Course" at 24ORE
Business School.
Nicola Magaletti, Business development manager with a degree in mechanical
engineering with over 30 years of work experience in structured companies, for which
he works in the management of innovation processes and in the launch of new business
initiatives in the industrial consultancy sector. Since 2018 he has been part of the Lum
Enterprise team as Operational Manager and Technical-Scientific Manager of the
“Smart District 4.0” R&D project of which the company is the lead.
