One of the major objectives of the military is to achieve high reliability and operational availability of their assets. Significant operational benefits are to be gained by employing highly reliable military assets. The major reasons why achieving high reliability is so important are: (1) maintaining weapons systems consumes a significant portion of the total defense budget; (2) mission reliability is a key factor in determining system effectiveness; (3) the sustainable level of system readiness is in large measure determined by its reliability and maintainability characteristics.
Operational availability is defined as the ratio of the time the system was available for operation to the total mission time. In accordance with Reliability Analysis Centre, Operational Availability is not just a function of design but also of maintenance policy, the logistics system and other supportability factors. It can be improved by improving the design, improving the support, or both. As availability is a measure of maintenance performance any effort resulting in an increase of ship operational availability is commendable. Na et al. derived with the key concept of the presented research is that availability can be simply expressed as uptime and can be formulated as “One minus Downtime”. Basically, the lower the downtime, the higher the availability.
There are two approaches for achieving and maintaining high operational availability of military systems: simultaneous replacement of mission-critical parts and prognostics asset-management strategies. The strategy of simultaneous replacement of mission-critical parts is, as its name suggests, the periodic, complete, and simultaneous replacement of only those parts that cause the system (or vehicle) to be inoperable if one or more of them fail.
The simultaneous replacement of all mission-critical parts increases operational availability over a specified period of time. It can be done at regular intervals or just before an upcoming mission to increase the system pulse reliability. Frequent simultaneous replacements as a maintenance strategy will lead to an increased average operational availability over extended time periods even for low-reliability systems, but it will also result in significant costs, the result of replacing most parts prematurely and underutilizing their service life. While the risk of failure is reduced because the platform now has new parts, the replaced parts are thrown away even though most of them still have some (in many cases substantial) useful life left.
The Difference Between Predictive Maintenance and Preventive Maintenance
Predictive maintenance (PdM) is maintenance that monitors the performance and condition of equipment during normal operation to reduce the likelihood of failures. Also known as condition-based maintenance, predictive maintenance has been utilized in the industrial world since the 1990s. The goal of predictive maintenance is the ability to first predict when equipment failure could occur (based on certain factors), followed by preventing the failure through regularly scheduled and corrective maintenance.
Predictive maintenance cannot exist without condition monitoring, which is defined as the continuous monitoring of machines during process conditions to ensure the optimal use of machines. There are three facets of condition monitoring: online, periodic and remote. Online condition monitoring is defined as the continuous monitoring of machines or production processes, with data collected on critical speeds and changing spindle positions. Periodic condition monitoring, which is achieved through vibration analysis, “gives insight into changing vibration behavior of installations” with a trend analysis. Lastly, remote condition monitoring, as its name suggests, allows equipment to be monitored from a remote location, with data transmitted for analysis.
While many maintenance programs use a bit of both, there are several differences between predictive maintenance and preventive maintenance. Preventive maintenance has involved inspecting and performing maintenance on machinery, regardless of whether the equipment was in need of maintenance. This maintenance schedule is based on either a usage or time trigger. For example, a heating unit is serviced every year before winter, or a car requires scheduled maintenance every 5,000 miles.
Also, preventive maintenance does not demand the condition monitoring component that predictive maintenance does. By not requiring condition monitoring, a preventive maintenance program does not involve as much capital investment in technology and training. Lastly, many preventive maintenance programs need manual data-gathering and analyzing. While preventive maintenance is determined by using the average life cycle of an asset, predictive maintenance is identified based on preset and predetermined conditions of specific pieces of equipment, utilizing different technologies. Predictive maintenance also requires more investments in people, training and equipment than preventive maintenance, but the time savings and cost savings will be greater in the long run.
The prognostics-based approach to maintenance is based on prediction of remaining life, parts vulnerable to failure are replaced just before they fail or before an upcoming mission. The prognostics approach is a more effective way to maintain desired operational availability. It allows reduced administrative and logistic delay times by anticipating upcoming failures and preparing the necessary parts in advance. In addition, the prognostics capability allows for intelligent maintenance—replacing only those parts whose remaining lifetime has reached a critical value. The prognostics approach, therefore, allows full utilization of each part’s service life; this operational availability increase is obtained at significantly lower cost (number of spares) than that of the frequent simultaneous-replacement maintenance strategy.
One of the main operational benefits of the prognostics approach is that it leads to potentially failure-free missions because it allows field commanders to select only those platforms whose operational availability exceeds the duration of the upcoming mission.
Three basic prognostics strategies can be distinguished: the traditional risk-based approach, failure-precursor-based approach, and physics-of-failure approach. The risk-based approach is based on the service life distribution of a part (assumed or known from its operational history). When the risk of operational failure reaches a typically very small number, the system operation is stopped, and the part is replaced. (This approach is routinely used on aircraft for life-critical parts.)
The risk of failure in this case describes a probability density distribution of the damage reaching critical size. Such a curve can, for example, describe a distribution of possible crack sizes at a certain location of a disc or blade in a turbine jet engine after a number of flights. As the number of flights increases, the distribution of possible crack sizes changes, reflecting increased probability of larger cracks. Eventually, there is a small but finite probability that the crack has reached a critical length, which is determined by material properties and upcoming loading conditions. The inspection interval times are set so that the parts are inspected just before the crack reaches critical size. The part is replaced in the event a defect is found or the part had flown a certain number of cycles, even if there is no defect found.
Precursor-based approach is applicable to situations when there is a degradation mechanism that can be observed and detected by a sensor. At some point during the vehicle operation, the damage-detection threshold is reached, indicating that damage has reached some level detectable by a sensor. In cases when this precursor time (i.e., the time between the progression of the damage from detectable to critical level) is long enough to plan ahead for the maintenance action, this is a useful approach. But in cases where a failure is detected just before it happens, the practical applicability is limited.
In Physics-of-failure approach, it is assumed that structural loads are continuously monitored, and damage progression is modeled by a physics-of-failure model and validated by on- or off-line sensors. The physics-of-failure model predicts the state of damage and the remaining life for an assumed loading during an upcoming mission.
Naval Data Aggregation and Planning with Probabilistic Reasoning (SNAPPR).
Because naval operators are responsible for performing preventative maintenance and monitoring the Operational Availability (AO) of a large number of vessel components within complex systems, there is a high risk of unnecessary or unwarranted maintenance of components, which can inflate costs and significantly delay deployments.
Although each vessel emits copious amounts of data, this data is confined to isolated storage within each vessel, and is only analyzed when on shore. A system that fuses these disparate data sources to monitor and predict each components AO in real-time can provide a comprehensive picture of fleet-wide (i.e., battlegroup) AO.
By generating a prioritized maintenance schedule based on this predicted AO, the system can also significantly reduce the cost of unnecessary maintenance. To address these challenges, we propose to design a System for Naval Data Aggregation and Planning with Probabilistic Reasoning (SNAPPR). SNAPPR uses a mature real-time data aggregation framework that fuses and reasons about data using robust probabilistic models to monitor and predict AO. SNAPPR then uses these predictions to generate an optimized preventative maintenance schedule (PMS) for the naval operator.
Defense Department to Extend Predictive Maintenance to Navy
A greater understanding of a machine’s life and health is critical for the military. Predictive maintenance enabled through small IoT devices will help track the health of specific parts of a ship. The Defense Innovation Unit (DIU) intends to bring predictive maintenance to the Navy, aimed at keeping naval, aircraft and ground vehicles online and avoiding costly last-minute repairs. Artificial intelligence has already been deployed in other areas of the U.S. Army, and the DIU is actively working to have predictive maintenance added to the Air Force fleet as well.
Speaking at the Dell Technologies Forum on Real Transformation, DUI Director of Strategic Engagement Mike Madsen said that they were in “very advanced discussions” with the Navy. The organization believes it will receive a contract next year to build the AI for the Navy. A greater understanding of a machine’s life and health is critical for the military, as adversaries begin to probe new ways of breaking through the country’s defenses. The predictive maintenance may be enabled through small IoT devices, which track the health of a specific part of the ship. The AI would have an overview of the entire operation, with the ability to quickly spot any abnormalities. While it is likely that most of the repair and decision work would remain with Navy officers, having an early warning system could avoid expensive repairs or failures in battle.
The Navy already deploys a wide range of sensors to check things like reduction gears, turbines, generators, and air conditioning plants. The DUI wants to expand that data collection, to include an overarching system that can notify Navy officers of issues. “The performance of any given asset is something we want to hold close. So I think what you have to do is you have to architect this from kind of the get-go with that kind of security mindset in mind,” said Rear Adm. Lorin Selby, a chief engineer in the Navy. “You can harvest that data and you could potentially discover vulnerabilities, so you have to protect that. That’s part of my project: as I do this, we’re bringing that security aspect into the program.”
Charles River Analytics Develops Advanced, Probabilistic Reasoning for System Components Onboard US Naval Vessels
Charles River Analytics Inc., developer of intelligent systems solutions, has received additional funding from the US Navy to build a System for Naval Data Aggregation and Planning with Probabilistic Reasoning (SNAPPR). SNAPPR creates probabilistic models of Naval system components, the environment in which they operate, and the missions that they serve, to help operators understand the Operational Availability (Ao) of hardware components onboard a Naval vessel. We have partnered with Raytheon IDS for the three-year SNAPPR follow-on contract, which is valued at nearly $1.5 million.
“AO is crucial to a mission’s success,” said Kenny Lu, Scientist at Charles River Analytics and Technical Lead on our SNAPPR effort. “Current Ao hardware protocols can result in unnecessary maintenance, causing crippling delays in mission-critical functions. SNAPPR’s probabilistic model produces real-time reasoning, estimates, and predictions of Ao to generate an optimized preventative maintenance schedule.”
According to CRA information, SNAPPR is aimed at helping operators understand the Operational Availability (Ao) of hardware components onboard naval vessels. SNAPPR furthers our growing portfolio of efforts in predictive health maintenance. Under the EMC2 and POWERED efforts, we used probabilistic programming to help predict resource gaps and faults in military operations systems. These efforts are powered by Figaro™, our open-source, probabilistic programming language for probabilistic modeling.
CRA scientist Kenny Lu, technical lead on the SNAPPR effort, said of the project: “Ao is crucial to a mission’s success. Current Ao hardware protocols can result in unnecessary maintenance, causing crippling delays in mission-critical functions. SNAPPR’s probabilistic model produces real-time reasoning, estimates, and predictions of Ao to generate an optimized preventative maintenance schedule.”
Navy Program Will Use AI on Drone Images to Predict Fleet Maintenance Needs
The Navy has been using drones to inspect the maintenance needs of its fleet and is getting ready to add some artificial intelligence and machine learning in the mix to help it prioritize repairs and predict future needs. In August 2020, Google Cloud and Simple Technology Solutions, or STS, announced an award through the Navy’s Small Business Innovation Research program to add predictive analytics to the branch’s maintenance program. Once up and running, the AI tool will use images taken by inspection drones to identify maintenance needs—particularly rust and corrosion—and prioritize the most pressing repairs. Eventually, the tool will be expected to predict future maintenance needs, as well.
The U.S. Navy currently spends billions per year on maintaining and repairing its fleet of vessels and other platforms like aircraft and facilities. It is a largely manual, labor-intensive process. STS will train Google Cloud AI and ML models on tens of thousands of images to identify corrosion as part of the first phase with the Navy. Using imagery and Google AI/ML technologies, STS will seek to drastically reduce the labor burden and safety risk associated with maintenance inspections.
For Phase I of the project, STS will use the Google Cloud AutoML tool to build a machine learning model trained on unclassified corrosion data from Navy inspection drone data, as well as public sources. Navy corrosion experts will work hand-in-hand with STS engineers to properly label images as they are ingested to ensure the model is trained accurately. STS will iteratively train and validate the model using custom inspection drone flight data, which will be uploaded using Google Cloud Storage for processing. The models will continually improve and update themselves based on the newly ingested data. “The ultimate goal, however, is to move from detection to prediction by expanding the subjects and sensors, and eventually integrating with Navy systems,” STS Chief Technology Officer Aaron Kilinski said.
To get there, the team will likely incorporate “additional Navy assets, data sets, sensors and integration with Navy systems,” the spokesperson said. “This is about automation, saving time and money, and keeping inspectors out of harm’s way,” Kilinski said. The project was awarded through the SBIR program “due to the technology innovation and potential for commercialization,” the release states, citing the two main focuses of the program—getting cutting-edge research into the military and supporting the broader economy through technology transfer.
“The manual inspection of Navy ships and vessels is a time-intensive, costly process that can drive up costs and slow down deployment,” said Mike Daniels, vice president of global public sector for Google Cloud. “We’re proud to work with the U.S. Navy and empower them with Google Cloud technology to transform corrosion inspections for greater efficiency and safety.”
References and Resources also include: