Asset Intelligence System A Practical Approach for Transforming from Reactive … · 2018-10-18 ·...
Transcript of Asset Intelligence System A Practical Approach for Transforming from Reactive … · 2018-10-18 ·...
Asset Intelligence System: A Practical Approach for Transforming
from Reactive to Proactive Asset Performance Management
White Paper, MainTrain 2018 Conference
Dr. Ananth Seshan, Mr. Mrinal Chakravarty
5G Automatika Ltd.
203 Colonnade Road, Suite 202,
Ottawa, ON K2E 7K3.
+1-613-368-4809 | www.5gautomatika.com | [email protected]
July 2018
©2018. 5G Automatika Ltd. All Rights Reserved.
2
Table of Contents
1. Introduction ................................................................................................................................................. 3
2. Strategic Importance of APM ....................................................................................................................... 4
3. Challenges .................................................................................................................................................... 4
4. Opportunity: Real-Time Asset Intelligence .................................................................................................. 6
5. Attributes of an Asset Intelligence System .................................................................................................. 6
6. Practical Opportunities to Improve Asset Performance using Asset Intelligence ....................................... 8
6.1 Opportunity 1: Removal of Non-Value-Added-Labour in Asset Usage Updates ....................................... 8
6.2 Opportunity 2: Asset Usage-Based Preventive Maintenance .................................................................... 9
6.3 Opportunity 3: Asset Condition-Based Predictive Maintenance ............................................................. 10
6.4 Opportunity 4: Trends or Rates of Change in Asset Performance ........................................................... 12
7. Conclusions ................................................................................................................................................ 13
8. References ................................................................................................................................................. 15
©2018. 5G Automatika Ltd. All Rights Reserved.
3
1. Introduction
With the advent of internet, digital and AI-based learning techniques, the ability to effect proactive
Asset Performance Management (APM) has become a practical reality. Be it mining, oil and gas,
water or power utilities, or process and discrete manufacturing industries, major players are
embarking on using real-time intelligence to make informed proactive decisions using a universally
integrated IT infrastructure.
Such a transformation from a reactive regime of APM to a proactive one has its own set of
challenges. While e-maintenance, AI-based predictive techniques and decision support systems
have been in place for over a decade and a half [1-10], the practical implementations of such
systems are not yet widespread due to various reasons (further outlined in section 3), including, the
need to leverage on existing legacy infrastructure in manufacturing organizations that do not
generally lend themselves to the acquisition of raw data in a standard and straightforward manner.
A practical “logical rules engine” based approach is presented in this paper to initiate the
transformation from a reactive to a proactive regime in APM in the presence of disparate systems
including, but not limited to, existing legacy infrastructure. Such an approach could subsequently
expand to more advanced and sophisticated implementation of AI-based deep learning algorithms.
We shall introduce the concept of the Asset Intelligence System that comprises the logical rules
engine as an intermediary orchestrator between plant floor and enterprise systems to facilitate
proactive APM.
In other words, this paper shall discuss the challenges and opportunities for making the
transformation using an Asset Intelligence System. We shall also showcase some best practices for
the practical implementation of such a system in the industry by means of real use cases.
©2018. 5G Automatika Ltd. All Rights Reserved.
4
2. Strategic Importance of APM
The main objectives of a well thought out APM program have been to maximize the availability and
reliability of assets. When these two metrics are improved, the return on (production) assets is
maximized. Given the fact that a major investment of the manufacturing (or utility) industry is on
production assets, achieving a better return on assets increases shareholder value - and is therefore
a strategic goal for most large enterprises.
Today, given the advancement in the state-of-the-art, it is possible to automatically capture
machine (production asset) data, transform the raw machine data to meta-data, discover
actionable intelligence from the meta-data and generate proactive maintenance actions to
maintain continuous and superior asset performance. Therefore, “opportunistic” maintenance
interventions can be generated in response to the “emerging” real-time intelligence on the
condition of an asset in the plant floor. This capability is useful beyond just maintaining operational
performance. In many instances, superior asset performance can result in a competitive advantage,
and conversely, asset degradation can relate to regulatory non-compliance, a safety risk, and/or, a
quality failure leading to a major financial loss. Therefore, APM is now approached by the industry
with strategic interest.
3. Challenges
Even though leaders in the industry are taking advantage of the major advancements in the area of
digital technologies to achieve effectiveness in APM programs – for instance, by introducing
Predictive and/or Proactive Maintenance of their assets, manufacturing companies by and large are
still practicing a predominantly reactive culture. That is, most large organizations are still allowing
their critical assets to fail without being able to predict the conditions of failure in advance and
preventing them. These organizations at best perform Preventive Maintenance (PM) but it is based
on calendar time. In other words, shutdowns of lines are planned, say, every two weeks for PM as
an example, and this is done without any significant evidence on whether the said two weeks (in
this case) are a good or a bad estimate of the time between the PMs. Similarly, a significant cost is
incurred in Corrective Maintenance and Emergency Maintenance.
©2018. 5G Automatika Ltd. All Rights Reserved.
5
The reasons for this situation can be summarized as follows:
(a) Predictive actions or measures for preventing failures will require real-time intelligence on
the assets’ performance, which most organizations do not have.
(b) Such intelligence must be derived from raw data streams – and therefore, an infrastructure
to convert the plethora of raw plant data available into contextual intelligence, in real-time,
is required which most organizations have not put in place.
(c) Vertical bi-directional integration between plant floor IT systems (Historian/MES) and
Enterprise Asset Management (EAM) systems that allow for generating immediate actions
to prevent failures (predictive maintenance work orders) and record root causes in the EAM
systems directly from the plant floor are required. Such an integration has either been only
partial, unidirectional and/or not yet established in most large organizations.
(d) Most organizations have legacy systems that need to be integrated for data acquisition. The
functionality, design, operation and performance of the components of the legacy system
need to be understood in order to design the types of changes that will be required for
building object wrappers to each component - in order to make them open standard.
(e) Cost justification for the above-mentioned is not straightforward, requires a cross
disciplinary effort and is not widely held knowledge. In most cases, the justification is made
with incomplete data by teams from a single discipline and so is not convincing to the
management to favourably look at investments in this area. In other words, the total cost of
downtime, its impact on profitability and the return on physical assets is not fully
understood in the industry in quantitative terms, even though, there is a qualitative
understanding of the impact.
©2018. 5G Automatika Ltd. All Rights Reserved.
6
4. Opportunity: Real-Time Asset Intelligence
As stated earlier, in order to perform good and continuously effective APM, there is a need for
generating proactive real-time actions that are based on intelligence and not data. In today’s
manufacturing environment, according to some reports, about 15,000 terabytes of data can be
generated every day. This is a lot of data!
Most of the collected data, unless converted into meta-data and further into intelligence, is not
useful. Asset Intelligence Systems are a breed of software programs that enable the collection of
raw data from the plant assets, controllers and drives and convert them into actionable intelligence.
The conversion process will involve the selection of data to be converted, aggregation of data in
multiple dimensions and automatically deriving some useful conclusions about the “status” or
“condition” of assets. The output of an Asset Intelligence System is the automatic generation of
useful and timely maintenance actions in resident asset management (or CMMS) systems in order
to proactively respond to a certain emerging behavior of assets that could result in a failure.
5. Attributes of an Asset Intelligence System
The following attributes are required for an Asset Intelligence System:
(a) It must enable seamless vertical integration between the plant floor assets and the
Enterprise Asset Management (EAM) or other legacy systems in a standard open systems
manner (as stated above) based on reusable libraries and specific wrappers.
(b) It must be capable of acquiring real-time raw data directly from machines using open
standards by interfacing with machine controllers (DCS), Historians or MES systems.
©2018. 5G Automatika Ltd. All Rights Reserved.
7
(c) It should enable the continuous and real-time derivation of meta-data from raw data, based
on “contextual” information about the data.
(d) It must have the capability to represent logical conditions as a set of rules that can be
triggered should the logical condition be found true in real-time (real-time actionable
intelligence).
(e) It must be capable of predicting future probabilities of failure of machines based on
symptoms using deep learning algorithms.
Figure 1 below depicts the scope and functionality of such an Asset Intelligence System. The critical
element in this system is the automated workflows that get automatically enabled between plant
floor and enterprise IT systems based on real-time intelligence derived from the Asset Intelligence
System that resides in-between the two aforementioned IT systems.
Examples of such automated workflows are provided in Section 6.
Figure 1: The Transformation of Data into Actionable Intelligence and Actions
©2018. 5G Automatika Ltd. All Rights Reserved.
8
6. Practical Opportunities to Improve Asset Performance using Asset Intelligence
6.1 Opportunity 1: Removal of Non-Value-Added-Labour in Asset Usage Updates
Preventive Maintenance (PM) is normally performed by tracking “elapsed time” (time-based PMs),
run hours, and/or, the number of cycles (usage-based PMs). The problem with using “elapsed time”
or Time-Based PMs is that machines could be under-maintained or over-maintained. Usage-Based
PMs are therefore more preferred because they present an opportunity to conduct PM
interventions based on actual data on usage rather than elapsed time.
In order to perform usage-based PMs, the usage data has to be updated periodically (say, every
shift or day) in order to ensure that the resident EAM system is able to generate a PM work order
when the usage crosses a certain threshold. The recording of such data in the EAM system happens
in a counter called “meters”. Meters are instantiated for every asset whose usage data must be
updated.
Typically, meter updates are manually entered by a human operator once in formatted paper forms
inside the plant application, and subsequently entered (copied) manually using the data input
screens in the enterprise application. As shown in Figure 2 below, an Asset Intelligence System can
establish a seamless connectivity and this will totally eliminate manual intervention, which means
two major benefits:
• Removal of a Non-Value-Added Activity and Cost of such labor.
• Removal of Potential Errors due to Human Fallibility.
©2018. 5G Automatika Ltd. All Rights Reserved.
9
Figure 2: Automated Workflow for Updating Usage of Assets (“Meter Updates”)
Typically, this opportunity results in immediate cost savings by eliminating one, if not two, full-time,
non-value-added labor activities.
6.2 Opportunity 2: Asset Usage-Based Preventive Maintenance
As opposed to updating meters in an asset management system that in turn will trigger workflows
for usage-based maintenance actions when thresholds for PM are crossed, it may be useful at times
to generate maintenance work order actions directly when usage for an asset has crossed a limit.
This is made possible by aggregating the run times of an asset continuously using the Asset
Intelligence System. When the values of the run times cross a threshold value, a PM work order can
be automatically generated in a resident EAM system by the Asset Intelligence System.
The example shown below in Figure 3 is an illustration of this opportunity in a wastewater utility. In
this case, the Asset Intelligence System connects multiple lift pumps in the utility to the EAM
system and aggregates the usage of the lift pumps individually. When any of the lift pumps cross
the given threshold of 5000 hours, a PM work order is automatically generated in the EAM system.
©2018. 5G Automatika Ltd. All Rights Reserved.
10
Figure 3: Automated Generation of a PM Work Order Based on Pump Usage
This opportunity improves (a) effectiveness of PM; (b) enforces “adequate” asset maintenance as opposed to “over-maintenance” or “under maintenance” which are both costly; and (c) improves the efficiency of the PM workflow by eliminating human inputs or systems that are not essential.
6.3 Opportunity 3: Asset Condition-Based Predictive Maintenance
The advantage of having an Asset Intelligence System is that a logical evaluation of an asset
condition can be made in real-time based on its performance and proactively acted upon. In other
words, process variables such as vibration, temperature, pressure, etc., can be logically combined
to understand the behavior of the asset in a holistic manner. This logical assessment can be used to
generate appropriate maintenance actions to predictively prevent a failure or even degradation in
asset performance.
The example below in Figure 4 is an illustration of a press shop responding opportunistically and
proactively to an emerging failure condition using an Asset Intelligence System. This case shows two
©2018. 5G Automatika Ltd. All Rights Reserved.
11
process variables being monitored – spindle vibration and temperature. The logical condition
illustrated herein is both variables crossing a threshold and staying above the threshold for more
than (say) 5 minutes – which would be a symptom for failure of the presses. The Asset Intelligence
System can monitor the values of the two variables as well as the duration of time they exceed the
threshold, and can therefore generate a Predictive Maintenance work order if the duration of both
variables exceeds 5 minutes (in this case).
Figure 4: Predictive Maintenance Generated by Asset Intelligence (example) -
Triggers created for a work order when the spindle temperature AND spindle vibration increase
above threshold for more than 5 minutes in any of the 100 CNC presses.
Another potential example of Predictive Maintenance helping the reduction of waste is shown in
Figure 5. In this case, the Asset Intelligence System monitors the reject counts of the parts
produced by the assets and if they cross 2% in any 2-hour period during production, the Asset
Intelligence System generates a Predictive Maintenance work order. This is an illustration of how a
plant can proactively respond to a “current” situation and rectify the situation if possible as
opposed to a postmortem analysis after the shift. Typically, say, if the reject counts from a machine
is less than 1% and it is suddenly found to be 2% in a 2-hour period (by the Asset Intelligence
System), obviously something is going wrong and the tools and/or assets in question needs to be
checked. The Asset Intelligence System can enable such timely actions to prevent waste.
©2018. 5G Automatika Ltd. All Rights Reserved.
12
Figure 5: Generating a Predictive Maintenance Workflow when the Number of Reject Counts
in any 2-hour Production Cycle is more than 2%.
6.4 Opportunity 4: Trends or Rates of Change in Asset Performance
The trends or rates of change of a process variable denotes the ‘velocity” in which a variable is
changing – which is, in most instances, more important than the absolute value of the variable
itself. In the example below which is applied to a process industry, the Asset Intelligence System
monitors the rates of change of pump temperature or discharge pressure drop – and if any of these
two values are greater than 1.5 times their last 30-minute average, it generates a Predictive
Maintenance work order automatically, as illustrated in Figure 6.
©2018. 5G Automatika Ltd. All Rights Reserved.
13
Figure 6: Generating a Predictive Maintenance Workflow
based on “Rates of Change” of Multiple Process Variables.
7. Conclusions
APM needs to be a proactive activity in order to prevent unplanned downtime. Until recently, there
were limitations and challenges for implementing proactive APM. With the advent of digital
technologies, it is now possible to deploy intermediary orchestrating systems, such as Asset
Intelligence Systems, to link between disparate machines and IT systems to effect early detection of
failures and prevent them. In other words, such systems can perform predictive and preventive
(proactive) maintenance using a logic-based rules engine, and in parallel, collect relevant data (and
experience) to “learn” (using deep learning algorithms) to make more futuristic predictions of asset
failures.
©2018. 5G Automatika Ltd. All Rights Reserved.
14
Since the above-mentioned Asset Intelligence System will have the capability of generating
actionable intelligence and automated workflows, it will result in the following opportunities for the
customer:
(a) Significant reduction in unplanned downtimes;
(b) Significant reduction in failure rates (or improvement in reliability);
(c) Removal of non-value-added labor due to the integration of workflows and business
processes between plant and enterprise;
(d) Unified visibility of production and maintenance information – in fact, maintenance actions
are driven by the need for enhanced performance;
(e) Reduction in the maintenance spend year on year;
(f) Mitigation of regulatory risks; and
(g) Due to most of the above-mentioned, increased return on (production) assets.
©2018. 5G Automatika Ltd. All Rights Reserved.
15
8. References
1. Laura Swanson, “Linking maintenance strategies to performance”, Department of Management,
Southern Illinois University Edwardsville, Edwardsville, IL 62026-1100, USA, Int. J. Production Economics
70 (2001), pp. 237-244.
2. D.N.P. Murthy, A. Atrens, J.A. Eccleston (School of Physical Sciences), “Strategic Maintenance
Management”, Journal of Quality in Maintenance Engineering, Volume: 8 Issue: 4, 2002.
3. Mari Cruz Garcia, Miguel A Sans-Bobi, Javierdel Pico, “SIMAP: Intelligent System for Predictive
Maintenance: Application to the health condition monitoring of a wind turbine gearbox”, Computers in
Industry, Volume 57, Issue 6, August 2006, pp. 552-568.
4. R.C.M. Yam, P.W. Tse, L. Li, P. Tu, “Intelligent Predictive Decision Support System for Condition Based
Maintenance”, The International Journal of Advanced Manufacturing Technology, Volume 17, Issue 5,
Feb 2001, pp. 383-291.
5. M. Hosek, J. Krishnaswamy, J. Prochazka. “Intelligent Condition Monitoring and Fault Diagnostic System
for Predictive Maintenance”, US Patent 7,882,394, 2011.
6. R.B. Faiz, E.A. Edirisinghe. “Decision making for Predictive Maintenance in Asset Information
Management”, Interdisciplinary Journal of Information, Volume 4, 2009.
7. G.A. Susto, A. Shirru, S. Pampuri, “Machine Learning for Predictive Maintenance – A multiple classifier
approach”, IEEE Transactions on Industrial Informatics, Volume 11, Issue 3, 2015.
8. Jay Lee, Jun Ni, Dragan Djurdjanovic, Hai Qui, Haito Liao, “Intelligent Prognostics Tool and e-
maintenance”, Volume 57, Issue 6, August 2006.
©2018. 5G Automatika Ltd. All Rights Reserved.
16
9. K. Wang, “Intelligent Predictive Maintenance – Industry 4.0 Scenario”, WIT Transactions on Industrial
Sciences, 2016.
10. Cesar E. Bravo, Luigi Saputelli, Francklin Rivas, Anna G Perez, Michael Nickolaou, George Zangl, Neil D
Guzman, Shahab Dean Mohaghegh, Gustavo Nunez, “State of the Art of Artificial Intelligence and
Predictive Analytics in the E&P Industry”, Volume 19, Issue 4, 2014.