In-situ monitoring places thermal, optical, and geometric sensors inside a laser powder bed fusion (LPBF) or directed energy deposition (DED) build chamber to measure each layer during printing, generating per-layer pass/fail data in real engineering units rather than relying on post-build CT scans and machined tensile coupons [S1][S4].
The economic driver is qualification cost: a single LPBF coupon lot, CT scan set, and metallographic review can run tens of thousands of dollars and weeks of calendar time per part geometry, so aerospace and medical OEMs are pushing sensor fusion and closed-loop control to compress acceptance testing from a full per-part campaign into a per-layer data stream [S1][S7][S8].
Sensor Modalities Used in Metal AM Build Chambers
Three sensor families dominate metal AM in-situ work: high-speed short-wave infrared (SWIR) or photodiode melt-pool cameras sampling at 10–100 kHz, thermal imaging (MWIR/LWIR) cameras capturing each layer at 50–200 Hz, and optical/laser-line profilometers that reconstruct 3D surface topography of the previously solidified layer before the next recoat [S3][S4].
Phase3D's commercial system quantifies layer height and surface flatness in micrometres during printing, turning a visual anomaly stream into numeric tolerance data that can be trended against CAD section views [S1]. The MDPI 2025 review notes that 3D surface monitoring is still underused relative to thermal imaging, even though layer-roughness excursions are a leading indicator of lack-of-fusion porosity downstream [S3].
Eddy-current and ultrasonic probes are emerging for conductive alloys, but the MDPI review flags both as poor at sizing small or irregular flaws compared with optical profilometry or in-process CT [S3].
Defect Signatures Sensors Actually Catch
The defect taxonomy that in-situ monitoring reliably flags maps onto four failure modes: lack-of-fusion pores (low melt-pool energy, visible as short dwell times and under-filled voxels), keyholing pores (excess energy, visible as bright plume spikes and tall vapour columns), balling and recoater streaking (visible as profilometry height outliers), and delamination (visible as inter-layer temperature drops and acoustic emission bursts) [S3][S6].
On a brass LPBF program, a multi-partner consortium is qualifying an optical monitoring system that flags porosity-promoting melt-pool instabilities during selective laser melting, treating the optical signal as a real-time porosity proxy [S5]. For stainless and titanium aerospace parts, the same logic applies: short-wave IR pyrometry tied to a layer-by-layer thermal model lets a build engineer reject a layer at recoat rather than after HIP and CT [S1][S4].
From Sensor Data to Part Qualification
Qualification is where in-situ monitoring earns its keep: each layer generates a record of melt-pool statistics, peak temperature, cooldown rate, and measured surface height, and that record is written to a per-build digital thread that an acceptance authority can audit [S1][S7].
Phase3D's framing, supported by an April 2026 NASA SBIR award, is that "for decades, qualifying a 3D-printed part for spaceflight has meant months of destructive testing and CT scanning, an approach that does not scale"; the funded work targets real-time surface-height quantification so that an as-built layer record can substitute for a fraction of the post-build inspection burden [S8].
In practice this means an OEM no longer needs to machine and tensile-test every build plate; instead, it can correlate in-process sensor traces against a smaller set of witness coupons, using the metal powder lot certificate, the 3D scanner layer profile, and the build log as the primary evidence package. A 2024 review (2025-08) argues that in-situ monitoring and post-process inspection are complementary rather than competing: thermal and optical streams catch process drift, while CT and metallography still resolve rare anomalies the sensors miss [S7].
Closed-Loop Control and Adaptive Builds
Closed-loop control is the second half of the value story: when a melt-pool camera flags a low-energy signature, the controller can raise laser power or slow scan speed on the next vector; when profilometry flags a high spot, the recoater pass can be reissued before the next layer is fused [S3][S6].
Phase3D's 2025 work at Argonne and 2026 NASA-funded layer-height quantification are the clearest commercial examples of feeding a metrology-grade signal back into process parameters rather than just logging it [S1][S8]. A 2025 MDPI review consolidates the academic side, covering scan-strategy adaptation in LPBF, DED, extrusion, and material jetting, and tying 3D surface monitoring to autonomous zero-defect manufacturing roadmaps [S3].
Where In-Situ Monitoring Fits, and Where It Does Not
In-situ monitoring is well matched to high-value, low-volume programs (aerospace flight parts, medical implants, defence spares) where the per-build cost of a CT scan plus tensile coupon is a meaningful fraction of the part cost, and where layer-level traceability is contractually required [S1][S7][S8]. It is a poor fit for high-volume commodity production where a 1% scrap rate is cheaper than a $200k sensor stack per machine, and for alloys whose process window is so wide that melt-pool drift does not predict mechanical scatter [S7].
For new AM programs, the practical adoption ladder runs: (1) retrofit thermal cameras for trending, (2) add melt-pool imaging and profilometry for per-layer go/no-go, (3) wire closed-loop laser-power control for known defect signatures, (4) substitute in-process records for a defined fraction of destructive coupons in the metal material qualification package. The U.S. Navy's September 2026 release of PPD 802-8436658, which sets a streamlined technical pathway for substituting legacy cast and wrought parts with additively manufactured metallic components in submarine construction and repair, is exactly the regulatory climate where a tight in-situ record matters [S5].
Limits, Failure Modes, and Open Problems
Sensor data is only as useful as the model that interprets it: an uncalibrated thermal camera will flag a Ti-6Al-4V build as anomalous simply because emissivity was set for stainless, and a melt-pool threshold tuned on a virgin metal powder lot can pass a reused, higher-oxygen lot that produces sub-surface pores the optical sensor cannot see [S3][S7].
Sub-surface defects remain the hard case: optical and thermal sensors are surface-bound, so lack-of-fusion pores a few hundred micrometres below the layer are invisible until HIP, CT, or destructive sectioning. The MDPI review explicitly lists this as the open problem that drives ongoing work in acoustic emission, in-process X-ray, and multi-modal sensor fusion [S3]. A practical risk is treating the in-situ record as a replacement for CT when, in fact, it is best used as a screening filter that reduces the CT scanning load, not a stand-alone certificate of conformance [S2][S7].
Standards Landscape and What to Watch
No single international standard yet prescribes which in-situ sensor stack makes a part "qualified"; the field is being driven by OEM specifications (NASA, U.S. Navy, major primes) and by ASTM/ISO working groups on AM process qualification [S7][S8]. The September 2026 U.S. Navy PPD 802-8436658 standard, which sets a streamlined technical pathway for substituting AM metallic components in submarine work, is the clearest 2026 signal that procurement authorities are willing to accept documented AM process records, including sensor data, in lieu of legacy forging certifications [S5].
Trackable signals over the next two quarters: Phase3D's NASA-funded layer-height quantification milestones under the SBIR award [S8], ASTM F42 / ISO/TC 261 working drafts that name in-situ data in acceptance criteria, and the first OEM specifications that allow a per-build sensor log to substitute for a defined fraction of witness-coupon testing. For background on the metrology side, see the entry on vibration condition monitoring, which shares the sensor-fusion logic applied to AM thermal and acoustic streams.
Related analysis: Aluminum can sheet supply tightens as recycled content targets diverge in 2026.