RETURN TO SELECTED PROJECTS
TIER S·Active Research / v1.2 Prototype
FLAGSHIP CASE STUDYAI / MLAnomaly DetectionIntelligent SystemsEdge ML

Self-Aware Spacecraft: Telemetry Anomaly Detection & System Reasoning

Autonomous anomaly detection and root-cause reasoning across 1,200+ telemetry streams under extreme edge latency constraints.

Status
Active Research / v1.2 Prototype
Timeline
6 Months · Fall 2025 — Present
Role
Lead ML & Systems Architect
Primary Metric
64% False Alarm Reduction
01Context // Challenge

Problem Definition

Real-time Autonomous Anomaly Detection in Deep Space

Deep-space exploratory craft produce thousands of asynchronous sensor streams (thermal, electrical, propulsion, attitude control). When subsystem anomalies occur, light-speed signal delays of 20 to 45 minutes prevent ground controllers from intervening before catastrophic subsystem damage occurs.

Why This Problem Is Difficult

  • [1]Massive Cross-Sensor Correlated Noise: Sensors exhibit non-linear physical couplings (e.g., solar array angle affects battery bus voltage and internal reaction wheel thermal gradients simultaneously).
  • [2]Extreme Zero-Day Failure Modes: Real spaceflight anomalies are rarely identical to ground simulation profiles, making supervised classification prone to missed events.
  • [3]Draconian Edge Budgets: Space-grade radiation-hardened processors (e.g., BAE RAD750 or Vorago ARM Cortex-M0) operate at clock speeds below 200MHz with tight power budgets under 15W.
  • [4]Dynamic Operational Modes: Routine spacecraft operations (e.g., thruster firing, reaction wheel desaturation, antenna slewing) produce massive step-changes in telemetry that trigger naive threshold detectors as false alarms.
CONSTRAINT #1Inference latency upper-bound: ≤ 25ms per telemetry batch
CONSTRAINT #2Total RAM footprint limit: ≤ 128 MB volatile memory
CONSTRAINT #3Zero reliance on cloud or ground-station roundtrips during critical flight regimes
02Rationale // Purpose

Motivation

Conventional spacecraft monitoring relies on static red-line/yellow-line threshold rules programmed months before launch. In complex multi-subsystem cascades, static limits either trigger hundreds of nuisance false alarms or miss compounding drift anomalies until a hardware watchdog trips.

Target Real-World Impact

Building an on-board reasoning model that learns the physical topology of interconnected spacecraft components enables early warning minutes before irreversible thermal or electrical runaway, preserving multi-billion-dollar planetary science missions.

03Architecture // Strategy

Approach & Hypothesis

A Dynamic Temporal Graph Neural Network (T-GNN) paired with an edge-optimized contrastive prediction head. The model represents telemetry sensors as nodes in an evolving dependency graph, predicting future continuous state distributions and flagging deviations in topological correlation rather than single-channel amplitudes.

Core Working Hypothesis

"Anomalies in complex cyber-physical machinery reveal themselves first as breakdowns in mutual information between physically coupled subsystems before exhibiting extreme numerical outliers."

TENET 01

Graph Structure Learning

Learn sensor dependency matrices dynamically via cosine similarity between projected temporal embeddings rather than static hand-drawn schematics.

TENET 02

Streaming Latency Guarantee

Decouple spatial graph aggregation from recurrent temporal updates so that per-tick inference runs as a fixed-size vector dot product.

TENET 03

Transparent Causal Attribution

Trace back high-loss nodes through their local graph neighborhoods to report the specific physical origin component, giving operators actionable root-cause diagnostics.

04Pipeline // Data Flow

System Architecture

Multi-stage execution flow from raw sensor streams down to root-cause attribution.

SYSTEM ARCHITECTURE DIAGRAM // v1.2
STAGES: 5 · BUS: ZERO-COPY SHARED MEMORY

End-to-end telemetry ingestion, spatial-temporal graph reasoning, and causal anomaly attribution pipeline running in real-time.

Hardware & OS

01. Telemetry Ingestion Layer

High-speed ring buffer collecting 1,200+ channels via CAN / SpaceWire interfaces over zero-copy shared memory.

C++20 Ring BufferZero-Copy IPCeBPF Tracing
Signal Processing

02. Dynamic Normalizer & Quantizer

Moving-window rolling z-score computation with outlier clipping and INT8 vector quantization for edge execution.

SIMD AVX2/NEONRobust Rolling Stats
Neural Representation

03. Temporal Graph Encoder

Learns time-varying edge weights between subsystem nodes and encodes multi-hop state dynamics using gated message passing.

Temporal GNNLinearized AttentionONNX Runtime
Inference & Diagnostics

04. Causal Anomaly Scoring Unit

Computes predictive error distribution per node and projects anomalous paths to attribute root cause to the originating subsystem.

Extreme Value Theory (EVT)Graph Path Attribution
Flight Action Engine

05. Autonomous Safety & Telemetry Sink

Emits structured event logs to spacecraft command bus and streams diagnostics over low-bandwidth downlink packets.

Protobuf SerializerFlight Bus Watchdog
Data Flow Pipelines
ingestpreprocess
1,200+ raw float channels @ 50Hz
preprocesstgn_encoder
Normalized INT8 window tensors
tgn_encoderreasoning
Latent node embeddings & predicted states
reasoningaction_bus
Attributed fault vectors & confidence scores
Component Topology Note:The ingestion layer feeds a lock-free circular buffer consumed by the preprocessing worker. The neural graph encoder operates on a fixed 30-tick temporal window. The scoring unit leverages Extreme Value Theory to adjust detection thresholds dynamically without ground recalibration.
05Foundations // Mathematics

Technical Deep Dive

Algorithms, mathematical representations, and low-level runtime optimizations.

The core innovation lies in fusing temporal representation learning with learned dynamic graph topology under bounded memory overhead.

05.1Dynamic Adjacency Matrix Formulation

Learning Subsystem Coupling Without Static Schematics
A_{ij}^{(t)} = \text{ReLU}\left( \tanh\left( \frac{e_i^{(t)} W_Q (e_j^{(t)} W_K)^T}{\sqrt{d}} \right) - \epsilon \right)

Rather than using a fixed manual adjacency graph, node embeddings e_i and e_j project into a shared subspace where temporal correlations form dynamic directed edges. The threshold epsilon ensures graph sparsity, limiting message-passing operations to top-k physically coupled neighbors and preventing quadratic computational blowup.

05.2Extreme Value Theory (EVT) Adaptive Thresholding

Non-Parametric Anomaly Scoring
P(X - \mu > x \mid X > \mu) \sim \left( 1 + \frac{\xi x}{\sigma} \right)^{-1/\xi}

Standard Gaussian assumption models fail because spacecraft telemetry anomalies reside in heavy-tailed distribution regimes. We fit a Generalized Pareto Distribution (GPD) over the prediction error tail above a high quantile mu, deriving mathematically grounded detection thresholds that adapt during spacecraft operational state changes.

05.3Low-Precision Quantization & Latency Profiling

Post-Training INT8 Execution on Edge Hardware

Exported PyTorch weights were quantized to INT8 with symmetric per-channel weight scaling and per-tensor activation ranges calibrated against simulated mission phases. Memory-aligned SIMD vector operations yielded a 3.4x speedup with less than 0.8% loss in anomaly F1-score.

06Bottleneck // Critical Path

The Hard Part

The single problem that demanded the most algorithmic reasoning and systems profiling.

Primary Core Bottleneck

Distinguishing Routine Operational Mode Switches from True Subsystem Degenerations

Why Standard Baseline Approaches Failed:

Standard baseline models (such as autoencoders and LSTMs) produced massive false-alarm bursts whenever the spacecraft fired thrusters for attitude adjustment or turned on power amplifiers. The sudden shift in sensor telemetry looked identical to a catastrophic fault to naive reconstruction loss algorithms.

Resolution Mechanism:

Implemented a conditioning subsystem context vector representing planned spacecraft commands. By feeding known actuator intent into the temporal prior, the graph network learns that sudden spikes in battery discharge coupled with thruster solenoid activation are expected system states, silencing nuisance alarms while immediately flagging unexpected cross-system deviations.

07Verification // Hypotheses

Experiments & Empirical Tests

Testing hypotheses through empirical sweeps, ablation baselines, and synthetic fault injections.

Conducted rigorous empirical benchmarking across the NASA Soil Moisture Active Passive (SMAP) and Mars Science Laboratory (MSL) telemetry datasets, as well as 48 hours of simulated multi-subsystem fault injection runs.

EXPERIMENT #1Verdict: Confirmed
Hypothesis:

"Dynamic Temporal Graph Networks outperform standard Multivariate LSTM Autoencoders in multi-channel anomaly detection accuracy."

Setup:Evaluated precision, recall, and F1-score on 82 labeled anomaly sequences in NASA SMAP with identical train/test splits.
Observation:T-GNN achieved an F1-score of 0.942 compared to 0.781 for LSTM-VAE, driven by significantly fewer false alarms during complex multi-channel transitions.
EXPERIMENT #2Verdict: Confirmed
Hypothesis:

"Graph sparsity thresholding allows linear scaling without degrading anomaly localization."

Setup:Swept epsilon from 0.0 to 0.8, measuring per-tick CPU cycles and root-cause localization accuracy across 500 injected sensor faults.
Observation:Setting epsilon = 0.3 reduced edge density by 74% and inference time by 61% while maintaining 96.8% root-cause top-3 attribution accuracy.
EXPERIMENT #3Verdict: Confirmed
Hypothesis:

"INT8 quantization maintains acceptable precision without retraining."

Setup:Compared FP32 model predictions against INT8 quantized model using TensorRT and ONNX Runtime on ARM Cortex-A72.
Observation:INT8 model incurred only a 0.007 drop in F1-score while slashing RAM utilization from 240MB down to 68MB.
08Benchmarks // Quantitative

Results & Metrics

Quantitative performance across accuracy, execution latency, and resource footprint.

The system demonstrated decisive improvements across accuracy, latency, and operational false-alarm resistance when measured against established aerospace baselines.

Overall F1-Score
0.942
Benchmark on NASA SMAP/MSL
Inference Latency
11.4 ms
Single-batch on edge ARM CPU
False Alarm Reduction
64%
Compared to static red-line rules
Memory Footprint
68 MB
Fits within 128MB budget
Baseline Benchmark Comparisons
Evaluation MetricBaseline ArchitectureOur T-GNN SystemDelta
Detection F1-Score0.781 (LSTM-VAE)0.942 (T-GNN)+20.6%
Inference Latency42.8 ms11.4 ms-73.3%
RAM Consumption260 MB68 MB-73.8%
Mean Time to Detect (MTTD)14.2 sec2.8 sec-80.2%
Root-Cause Attribution (Top-3)44.0%96.8%+120.0%
09Autopsy // Honesty

The Failure Principle

Demonstrating engineering maturity through explicit failure diagnosis and systematic redesign.

Development was not a linear path of instant success. The initial implementation suffered a severe architectural bottleneck that required tearing down the first attention design.

01. Attempt

Built a full all-to-all cross-attention Transformer across 1,200 sensor channels to model telemetry correlations directly.

02. Failure

The model caused continuous memory allocation faults and took over 380ms per telemetry tick on the edge testbed — far exceeding the 25ms hard real-time limit.

03. Diagnosis

Profiled the memory allocator using Linux perf and eBPF. The O(N^2) attention matrix created devastating cache thrashing and memory bus saturation on resource-constrained ARM architectures.

04. Modification

Scrapped full dense self-attention in favor of a sparse Temporal Graph Network with dynamic k-NN neighborhood pruning and linear spatial message passing.

05. Result

Reduced per-tick inference latency from 380ms to 11.4ms (a 33x acceleration) while fitting within the 68MB resident memory footprint.

10Lineage // Versioning

System Evolution

The progression from rudimentary statistical baselines to edge-quantized streaming inference.

v0.1·Month 1

Heuristic & Statistical Baseline

Architecture:

Rolling z-scores and static threshold bands implemented in Python scripts.

Outcome:

Established baseline metrics; proved static thresholds fail miserably on coupled non-linear sensors.

Key Limitation:

Generated over 180 false alarms per 1,000 operational hours.

v0.8·Months 2–3

Recurrent Autoencoder (LSTM-VAE)

Architecture:

Unsupervised temporal reconstruction error using multi-layer LSTM autoencoders.

Outcome:

Successfully flagged novel anomalies without labels, achieving 0.781 F1.

Key Limitation:

Severe false alarm spikes during planned mode switches; high inference latency.

v1.0·Months 4–5

Temporal Graph Neural Network

Architecture:

Graph structure learning + spatial-temporal message passing + EVT thresholding.

Outcome:

Achieved 0.942 F1 and enabled precise root-cause attribution to specific faulty sensors.

Key Limitation:

Floating-point FP32 weights consumed 240MB RAM, exceeding target satellite flight computer limits.

v1.2 (Current)·Month 6

Edge-Quantized C++ Runtime

Architecture:

INT8 quantized engine in C++20 with ZeroMQ ring buffer and ONNX Runtime execution.

Outcome:

Inference latency dropped to 11.4ms with 68MB RAM footprint, verified on ARM edge testbed.

Key Limitation:

Still requires offline graph pre-training before deployment.

11Takeaways // Knowledge

Engineering Lessons

Machine Learning Engineering

  • Systems constraints dictate ML architectures: An algorithm that achieves state-of-the-art accuracy on an A100 GPU is useless if it cannot meet the clock cycle and cache footprint of the target deployment processor.
  • Ablation studies save weeks of wasted effort: Decoupling the graph spatial encoder from the temporal aggregator early on proved that 80% of accuracy came from local physical sensor clustering, not distant cross-system attention.
  • Telemetry is never clean: Sensor dropout, clock drift between subsystems, and packet jitter must be engineered into the synthetic training simulator from day one.

Systems & Infrastructure

  • Zero-copy memory management is non-negotiable: Eliminating JSON serialization and memcpy steps between the ingestion bus and tensor input reduced end-to-end latency by 18ms alone.
  • Fail-safe fallbacks are mandatory in mission-critical systems: The ML engine runs as an advisory copilot; traditional hardware threshold watchdogs remain active as the final safety circuit.
12Roadmap // Horizon

Future Work

Upcoming Milestones

  • 01.Hardware-in-the-loop (HIL) testing on an active spacecraft bus simulator with real radiation-tolerant flight computers.
  • 02.Continual on-device self-supervised adaptation using low-rank adapter updates (LoRA) during multi-year cruise phases.
  • 03.Integration of neuromorphic event-sensor telemetry for micro-vibration and structural flexure monitoring.

Open Research Questions

  • ?Can causal graph discovery prove invariant bounds under unmodeled space weather radiation events?
  • ?How can we formally verify that neural anomaly detectors never suppress critical safety alarms during emergency safe-mode entries?
Ecosystem // Tooling

Complete Technical Stack

Core Machine Learning
PyTorch 2.4Temporal Graph Networks (TGN)Quantized INT8 PyTorchContrastive Representation Learning
Systems & Inference
C++20 EngineONNX Runtime EdgeZeroMQ Telemetry BusLinux eBPF TracingSIMD Intrinsics
Telemetry & Benchmarking
NASA SMAP / MSL DatasetsHigh-Rate Synthetic Fault InjectorPrometheusGrafana Telemetry Sink