TECHNOLOGY AREA(S): Information Systems

ACQUISITION PROGRAM: Robust Combat Power Control FNC

OBJECTIVE: Develop machine learning and data analytics methods that enhance state and situational awareness (SA) for shipboard machinery systems.

DESCRIPTION: The Navy is planning to develop, qualify, supply, and maintain a standard, Navy-owned, software baseline for surface ship machinery control and condition monitoring systems (MCS).� Enhancements to the MCS baseline shall be capable of running on a JAVA virtual machine. This MCS software baseline will have a common look and feel, operating system, algorithmic infrastructure, application interfaces, and qualification and transition process for new software modules, across all future delivered surface platforms.� This will include robust cybersecurity assurance.� The software products developed under this STTR topic are primarily algorithmic in nature (vice standalone products that are directly human-usable upon delivery), and as such will be incorporated and integrated directly into the algorithmic base of the standardized MCS baseline, using all the pre-established guidelines, processes, and procedures thereof.

Naval vessels are replete with state-of-the-art sensors that monitor vital ship and auxiliary system functions.� Battle situations or other circumstances that create rapid loss of offensive, operational, communication, and auxiliary capacity present a series of challenges to operators tasked with rapidly restoring capabilities.� Lack of prioritization of the importance and larger contextual meaning of multiple alerts can prevent the operator from rapidly reclaiming optimum state awareness for both the human operator and the autonomous control system.� New tools and technologies are required to support current alerts and control systems and help improve situational certainty so that best-fit responses are applied as soon as possible.� Major gains have been made in autonomy and Augmented Intelligence (AI), but additional work is required for both automated response and operator situational awareness to produce faster and better decisions.� Advances in data science, data fusion, and machine learning show promise in effective management of continuous streams of data to supplement current control systems, producing higher informational accuracy in less time and getting the right data to operators when needed to produce more optimal crisis responses.� The use of cognitive technology and machine learning can process sensor data rapidly and analyze events, find patterns, paths, and options for configuring auxiliary system capabilities with much greater speed and agility than those reliant solely on human interpretation and decision making.� Innovative methodology will provide real-time decision tools, as the integration of high-energy weapons and sensors requires the machinery systems to coordinate the delivery of power and cooling resources in an optimal manner to support sustained combat operations.� This advanced state awareness methodology must be compatible with control methods being developed under the Robust Combat Power Control Future Naval Capability Program.� It is necessary to communicate intent to the machinery control system, present options for solutions to plant alignments and resource allocations, maintain SA even in the event of battle damage, and present the right alarms and information in a manner that does not overload the human operator but rather elicits information from the human in a fashion complementary to the machine intelligence.

This topic seeks to use various data fusion technologies for analysis of unstructured data (text, images, etc.) and structured (signal feeds, database items, etc.) information to make determinations and useful observations around the context of the information combining data sets for automated decision support and predictive capabilities.� Information collected from equipment and sensors are expected to be combined with other contextual information to provide more advanced predictive models and recommended actions.� Leveraging historical situational data, best practice scenarios, and other data collected from past events can help to provide operators with valuable insights and recommendations for potential actions not previously available.� The software products developed are primarily algorithmic in nature (vice standalone products that are directly human-usable upon delivery), and as such will be incorporated and integrated directly into the algorithmic base of the standardized MCS baseline, using all the pre-established guidelines, processes, and procedures.

A machine learning system can also provide a pathway to better item-specific predictive maintenance based upon analytics and learning over time.� The cognitive system that this topic addresses could also use data collected from a wide range of venues to develop predictive maintenance schedules based on patterns, and actual utilization of specific critical components like generators and pumps.

The solution sought will develop a user interface that allows an operator view data quickly from numerous components alongside suggestions and patterns based upon past data or reoccurring incidents from related components to allow them to provide prioritized recommended actions.� The goal is not to replace human interaction and decision-making, rather it is to support the operator by leveraging AI technologies that combine data, locate patterns more rapidly, and provide operators with a more comprehensive view of their present situation.

PHASE I: Develop a concept for Situational Awareness for Mission Critical Ship Systems for Naval Applications that meet the requirements described above.� The small business will demonstrate the feasibility of the concept in meeting Navy needs and will establish that the concept can be developed into a useful product for the Navy.� Feasibility will be established by component evaluation and analytical modeling.� The Phase I option, if awarded, will include the initial layout and specifications to build the prototype in Phase II. Develop a Phase II plan.

PHASE II: Based on the results of Phase I and the Phase II Statement of Work (SOW), the small business will develop a prototype for evaluation. The prototype will be evaluated to determine its capability in meeting the performance goals defined in Phase II SOW and the Navy requirements for Situational Awareness for Mission Critical Ship Systems. System performance will be demonstrated through prototype evaluation and modeling or analytical methods over the required range of parameters including numerous deployment cycles.� Evaluation results will be used to refine the prototype into an initial design that will meet Navy requirements. The small business will assess integration and risk and develop a Software Development Plan (SDP). The small business will prepare a Phase III development plan to transition the technology to Navy use.

PHASE III DUAL USE APPLICATIONS: Support the Navy in transitioning the technology for Navy use.� The small business will develop Situational Awareness for Mission Critical Ship Systems according to the Phase II SOW for evaluation to determine its effectiveness in an operationally relevant environment.� The small business will support the Navy for test and validation to certify and qualify the system for Navy use.� The Navy is planning to develop, qualify, supply, and maintain a standard, Navy-owned, software baseline for surface ship machinery control and condition monitoring systems (MCS).� This MCS software baseline will have a common look and feel, operating system, algorithmic infrastructure, application interfaces, and qualification and transition process for new software modules, across all future delivered surface platforms.� This will include robust cybersecurity assurance.

Complex decision tools capable of alert prioritization and state awareness can provide significant advantages in the fields of medicine and finance.� Advanced cognitive controls are in demand in manufacturing, industrial facilities in hazardous environments, space, shipbuilding, and in system management of power plant and electric grid control.

REFERENCES:

1. LeCun, Yann, Yoshua Bengio & Geoffrey Hinton. �Deep Learning.� Nature International Weekly Journal of Science, Volume 521, Issue 7553, 27 May 2015. http://www.nature.com/nature/journal/v521/n7553/abs/nature14539.html

2. Ngiam, J., A. Khosla, M. Kim, J. Nam, H Lee, and A.Y. Ng. �Multimodal Deep Learning.� Proceedings of the 28th International Conference on Machine Learning, Bellevue, WA, USA, 2011. http://machinelearning.wustl.edu/mlpapers/paper_files/ICML2011Ngiam_399.pdf

3. Schmidhuber, J�rgen. �Deep Learning in Neural Networks: an Overview.� Neural Networks, Volume 61, January 2015, Pages 85�117. http://doi.org/10.1016/j.neunet.2014.09.003

KEYWORDS: Alert Prioritization; Deep Learning; Multimodal Learning; Informational Dosing; Augmented Intelligence; Data Fusion