Projects, Infrastructures, Programming - Recent & Summary of list of other experience
2020 - September to 2021
April - Present
Helping Dr. Alan Wagner restart an infrastructure useful in many vertical marketplaces. A high end distributed computing platform managed by a single config file defining services, nodes, data distribution of parallel processing computational graph, auto code generation, loaded and run with posted outputs back to user consumable client interfaces. Various application POC have been demonstrated in stock market, aviation, data centers, horse racing - predictive analytics.
I am relearning Alan's Specialties:
- MPI - Message Passing Infrastructure,
- FG-MPI - Fine Grained MPI
- CSP - Concurrent Sequential Processing,
- SCTP - Streaming Control Transmission Protocol
- Using Processes Not Threads
- Writing MPI Scheduler to generate optimal, cooperative process graph execution
- Running 100 Million Processes in Steps on Way to Exascale Processing
- Java Client - Original - Shows Correlated Stocks in Market
- Python Client to Elasticsearch
March, Sept, November-December 2020
Name: Independent Study
US State Covid Data Analysis & Visualization using Elasticsearch
Allen Institute Covid PubMed Abstract & Paper Analysis - more than 2000 unique fields
Allen Institute Covid PubMed Abstract - Galpin suggested search - using Elasticsearch MI illustrating partial topology of co-associated terms to help validate hypothesis of interest
September-October 2020
Name: NASA B747 Engine Flight Anomalies Demonstration, 500 Gb
Infrastructure: Elasticsearch, Google Compute Cluster, Linux,
Bash Scripts to:
(1) Partition Flight Data,
(2) Find Min, Max for all Params across all Flights,
(3) Run Python Code ; Generate Similarity Metric - 26 ^ 2 / 2 - since matrix is symmetric
(4) Push Similarity Metrics into Elasticsearch using Python Client API - Eland
(5) Push specific flight anomalies into Elasticsearch to validate recurrent degradation patterns
April 2018-September 2020 Boeing CH-47 Platform Group
Name: Lithium Ion Battery Thermal Runaway Modeling - Spring 2020
Working with Boeing Battery Enterprise Experts and CH-47 Platform Safety experts researched thermal runaways models including chemistry. Changed project trajectory going forward due to combination of toxicity and likely thermal runaway event due to battery management system using Julia Differential Equation package in review with Boeing Battery experts.
Name: UK MH-47D Flight Safety Study using Operational Data and Structural Usage Monitoring models.
Study illustrated beyond design limit flight data validated by Boeing UK Lead Engineer and Boeing Chief Engineer for MH-47D; non-longer utilized variant of US CH-47 platform. Systems under study included gearbox, fuel system, generators, electrical subsystem, engine torque, engine gas temperature, engine air bleed flow.
Structural Usage Monitoring (SUMS) system is an internationally recognized method:
http://www.tc.faa.gov/its/worldpac/techrpt/tc15-15.pdf
Standardized in the early 90's, SUMS utilizes extensive filtering and first order statistics: mean, variance and complex rule set.
Using Elasticsearch as the index, Similarity metrics calculated via external Java and separately Python programs to generate from 50 Gb of flight data that illustrated trends and occasional beyond design limits on design elements. Due to averaging SUMS didn't find limits exceeded.
Upshot is that UK recently starting buying newer models and Boeing received additional engine service support contracts.
Same analysis on US models resulted in similar findings.
Name: Teaching Elasticsearch to Senior Engineering Teams Safety and Reliability - Jan 2019-2020
Infrastructure is combination of Oracle Data Warehouse, Saffron Tech Memex, Elasticsearch to generate standard reliability metrics MTBF (mean time between failure), MTBUN ( mean time between unscheduled maintenance), MTBA ( mean time between aborted flights ), MTBAA (-air abort), MTBGA (ground abort) and RUL (remaining useful life - that is directly related to inventory). For more than 5 million maintenance events across 1000 components, subsystems and systems. The Saffron Tech Memex stored all the words from the logbook maintenance entries mapped to entities and attributes then imputed or inferred records containing blanks for various fields that human traditionally filled in by hand.
Additional correlations were associated with manufacturing, flight test data and operational flight data.
e.g. specific ATA-6 level component replaced as part of unscheduled maintenance due to an air abort
My roles were to create user documentation, including how to install ELK, load data, create visualization templates and dashboards, illustrate during classes then help users apply their common work packages within Elasticsearch rather than continue using Excel.
Name: Supporting Memex Infrastructure for the Engineering Teams Safety and Reliability - April 2018- Sept 2020
Saffron Memex is hosted on 5 linux clusters (8 core / 16 GB per node). Python and Java client code interrogate the index to classify/infer/impute predictive fields with 80-100% accuracy. The Saffron Tech Memex creates a compressed graph, co-associate large sparse matrix of all related entities and attributes across all data sources and data types.
Name: Due Diligence Activities for Scowal Investments - March 2018 - August 2019
Various projects related to credit card fraud and block chain.
Analatom - Handbuilt Memex 2015-March 2018
Memex Clusters - 3 - Dual GPU Linux Clusters using LLNL Fastbit to generate compressed graphs and similarity metrics. Using Julia, python and Elasticsearch for the index of indexes (containing compressed graphs).
Used in USAF C-130J working with USAF Chief Engineer, Propulsion Lead Engineer and MERC Integration team initially to characterize fuel system and hard landings. After resolving both of those I also found key to NFF - No Fault Found - anomalies - the actual phenomena requires system of systems analysis, as it is associated with aircraft control systems.
Used in ONR - Office of Naval Research - Additive Manufacturing - Reliability Study
Utilized C, C++, LAMMPS ( https://lammps.sandia.gov/ ), Similarity Metric Spaces, Python, Julia, SciKit
2007-2014 Boeing Enterprise Strategic Growth Group - Entity Analysis Lead At Boeing's Enterprise Strategic Growth group in past assignments one tech failed task, earthquake prediction. With sufficient data there is a team that now reliably predicts 7 of 10 earthquakes up to a week ahead. The benefit of earthquake prediction is that critical assets can be moved out of harms way and emergency management assets and inventory can be deployed nearby to help. I did succeed in hurricane, tornado track and weather forecasting of Pacific ocean using Memex of reported weather metrics, as opposed to WWM (Worldwide Weather Model).
In other technical tasks I have succeeded in finding precursors to failure in complex systems utilizing robust time series analysis methods. This required working with high caliber technical staff including chief engineers, integration engineers and customer TPOC/Technical Point of Contact; eventually meeting heads of USAF, Navy and US Army Integrated Vehicle Health Management.
The question of why I didn't succeed in Boeing on this problem is that despite best efforts with support of SVP / VP's, groups in charge of data repeatedly denied access. In the SBIR USAF contract working with chief and senior engineers, I was granted access to all data sources: OEM engine data set (with SHM/Structural Health Management parameters), OEM aircraft FDR/Flight Data Recorder, faults, MX/Maintenance Events, FIM/Fault Isolation Manual and THOPS/Theory of Operations. The C130J was built by Lockheed Martin with Rolls Royce engines.
Complex system degradation mapped into failure modes are aircraft and their subsystems electrical, propulsion, generators, hydraulics, flap / lift systems, fuel systems, batteries, battery management systems, thermal systems, air flow, vibration, radar, RF - Uhf, Vhf, Hf.
Elsewhere I discuss the Memex; technically the Memex is an index of indexes that are co-associated with a complex systems including dynamics. The Memex provides the set of parameters utilized as inputs to a set of methods. Using large sets of inputs, the methods define a set of similarity metric spaces. The outputs are either a known classification label assigned by a human or a generalized unsupervised similarity metric useful for clustering.
Many ML methods are counter productive in finding degradation due to systemic engineering / mathematics legacy thinking with regard to models. Thus I am at odds with current ML practices. I do continually review state-of-art methods in commercial and academic realms.
2003-2006 AWACS HCI, infrastructure and Comm group.
C, C++, Erlang, Java, JBoss, JGroups (DHT - HSQLDB HA), Ensemble (DHT-MySQL HA)
2001-2003 Terabeam - Free Space Optical Link - inner city 40Gbs ethernet links
Met Dr. Alan Wagner here; was able to learn from him advanced techniques in distributed computing practice preferring agent based message passing architectures, owning the scheduling and running on cores. Terabeam team built early implementations of MPLS, mobile network analysis agents, group communication protocols (originating from Cornell's Network Systems group that built the NYSE, Zurich Exchange and French Air Traffic Control System). Co-authored patent to Link Quality agent orchestrated fail over scenarios - actively reconfiguration of network elements with proper notification of network management administrative elements, including agent based human-like logins into Cisco switches and Routers in order to put into place fail-over configuration then putting back in the proper routing when optical network link recovered.
C, Cisco OS, C++, Distributed Network Agents, Java, JBoss, JGroups, Perl, SNMP, Linux VLAN
Sept 1999-2001 AT&T Fixed Wireless / (originally McCaw) - put cell phones onto side of houses
Supported devops engineering groups, system integration, test, production test and production upgrades into AT&T network, including auto-install and upgrade fail-over back-up previous versions.
C, CMIP, C++, Corba, NMS, NNS, WEMS, C - Cell and Base station elements - embedded
1991-1999 Industrial Systems bought out by AspenTech
Large distributed IOT industrial plants building supervisory control systems with up to 100K samples refreshed at 1, 5, 10, 30 Hz data stream elements, running control models and sending set-point targets to help optimize plants with builtin safety fail-over to "quiet" stable conditions.
C, C++, Oracle, Informix, Sybase, MS SQL, control systems - PID loops, Modbus, PLC, Scada
1988-1990 AT&T Bell Labs
Auto-reconfigured national network reducing service delivery time from 30 days to 10 minutes. Still working in national and wireless networks. Using rule based expert system to synchronize all the elements in the national network with engineering databases.
1986-1988 Bell Communications Research / Split from Bell Labs in 1984
In Statistical Mathematical Modeling group characterizing lasers, fiber, diodes and coupling elements.
Comments
Post a Comment