Leeham News and Analysis
There's more to real news than a news release.
Bjorn’s Corner: Faster aircraft development. Part 8. Conceptual System design.
September 19, 2025, ©. Leeham News: We do a series about ideas on how the long development times for large airliners can be shortened. New project talks about cutting development time and reaching certification and production faster than previous projects..
The series will discuss the typical development cycles for an FAA Part 25 aircraft, called a transport category aircraft, and what different ideas there are to reduce the development times.
We will use the Gantt plan in Figure 1 as a base for our discussions.
Figure 1. A generic new Part 25 airliner development plan. Source: Leeham Co. Click to see better.
*** Special thanks to Ron Everlove for helping with this article***
Conceptual System Design
During the Conceptual design phase, the team needs to mature the design of the entire aircraft, including structures, systems, interior, etc. For a 14 CFR Part 25 airliner, the design needs to comply with various systems requirements, such as Part 25.1309, for equipment, systems, and installation. The work to comply with such regulations begins at the Conceptual design phase.
The system team needs to identify several aircraft functions. Examples of requirements:
- provide required thrust during all phases of operation
- provide required directional control during all phases of operation
These may sound obvious, but they are essential for subsequent analyses and compliance determinations. The team needs to evaluate failure cases for each of these functions and determine their effects.
The “provide thrust” function needs to be broken down into various phases, such as taxi, takeoff, climb, cruise, etc. The team will need to look at failure cases such as uncommanded thrust increase, asymmetric thrust, one-engine-inoperable, etc. Some cases may cause a slight increase in workload for the pilots. Other cases may result in injuries or death. The team would assign hazard levels based on the effects of the failure cases.
Similarly, if the team examines the “provide directional control” function, it needs to evaluate the impact for each phase. We are skipping flight phases for now and focusing on taxi, takeoff, and landing. These phases may require more granularity for this analysis. If a plane loses directional control at a very low speed, it may not be a big problem. If the plane loses directional control at landing and veers off a runway, it could result in a fatal crash.
System Level Requirements
Armed with the functions, failure cases, and hazard levels, the systems team can start evaluating system-level functions and requirements to develop the architectures.
For the uncommanded thrust increase case, would a fan speed and throttle quadrant position synchronization be sufficient?
For the ground directional control, is a backup system needed for the nosewheel steering, or can we improve the reliability of certain nosewheel steering components to meet the failure probabilities needed?
Again, aircraft design is an iterative process, and the back-and-forth communication within a cross-functional team is absolutely key for getting to the design maturity needed at this phase and thus securing the project’s plans (now or for later development stages. Having to redo the previous stage analysis later to change a system architecture is very detrimental to the project).
Supplier Interaction
Depending on make/buy decisions for system parts, the team would also start collaborating with suppliers. Suppliers would present their solutions and provide recommendations. They could pitch off-the-shelf, modified, brand-new, or a combination of system ideas.
This is where supply chain, engineering, manufacturing, customer support, and program management need to review these proposals together to identify the best solution for the program. Performance, cost, schedule, and risk are the topics at this stage.
By the end of the Conceptual system design phase, the systems team should have a good understanding of all key systems and be able to agree on system architectures. These would then be subject to more detailed analyses in later program phases.
AI Support for Conceptual System design?
Can AI help speed up the process? We decided to put ChatGPT to the test again. This is the prompt we entered:
I’m designing a Boeing 737 and Airbus A320 replacement aircraft. It’ll use a hybrid propulsion system. Would you generate a list of aircraft functions for the safety analysis?
The following is the output:
Absolutely—here’s a comprehensive, system-oriented master list of aircraft functions you can use as a starting point for your Functional Hazard Assessment (FHA) and subsequent PSSA/SSA work. It’s tailored for a 737/A320-class single-aisle with hybrid propulsion. I’ve written functions in “provide/perform/ensure/monitor/protect” language to make them FHA-friendly and grouped them by system. Use across all phases (preflight, taxi, takeoff, climb, cruise, descent, approach/landing, go-around, diversion, abnormal/emergency).
0) Aircraft-Level & Operational
Provide safe flight and landing in all operating conditions and required dispatch configurations.
Provide continued safe flight and landing following single failures and specified multiple failures.
Provide performance to meet certified field lengths, climb gradients, obstacle clearance, and missed approach.
Provide ETOPS/LROPS capability (if targeted) including dispatch, en-route diversion, and cold-soak.
Provide crew procedures, checklists, alerts, and training cues for normal/abnormal/emergency ops.
Provide aircraft configuration management across all phases (flaps/slats/gear/bleeds/energy modes).
Provide environmental qualification and lightning/HIRF/EMC robustness for all systems.
Provide crashworthiness, occupant protection, and emergency egress.
1) Propulsion & Hybrid Energy Management
Provide required thrust across envelope (including hot/high, degraded modes).
Provide hybrid power blending (thermal engine + electric motor assist) to commanded thrust.
Provide engine start, stabilization, and shutdown (air/ground, normal/abnormal).
Provide motor-only taxi (if applicable) and low-emissions ground ops.
Provide mode management: ICE-only, electric-assist, electric-only, charge-sustain, charge-deplete, idle-charge, regen (if implemented).
Provide automatic energy optimization vs. mission phase, noise, emissions, and thermal limits.
Provide autonomous reversion between modes upon failures or limit exceedances.
Provide propulsor/motor torque control, synchronization, and surge/stall protection.
Provide engine/motor relight/restart capability within defined envelope.
Provide FADEC/EMDEC (engine + motor controller) command/monitoring, built-in test, and fault accommodation.
Provide thrust asymmetry management, automatic thrust reduction/shutoff upon hazardous conditions.
Provide vibration/overspeed/over-temperature/over-current protection for engines, motors, gearboxes, inverters.
Provide fire/overheat detection and extinguishing for nacelles, e-machines, power electronics, and battery bays.
Provide safe shutdown and electrical isolation for uncontained failure or fire.
Provide inlet/propulsor ice protection and FOD protection (incl. motor & inverter).
Provide thrust/torque limits to protect structures/shafts/gearboxes.
2) High-Voltage (HV) Energy Storage & Power Electronics (Hybrid-Specific)
Provide HV battery energy storage sized for mission use cases (assist, taxi, APU replacement, emergency).
Provide battery management system (BMS): cell balancing, SOC/SOH/SOHx estimation, current/voltage/temp limits.
Provide HV contactor/interlock management and pre-charge/discharge control.
Provide battery thermal management (heating/cooling), cold-soak recovery, hot-day margins.
Provide venting, off-gas detection, smoke detection, and explosion relief paths; prevent propagation.
Provide crash/ditch protection, shock/vibration isolation, and liquid ingress protection.
Provide HV distribution buses, DC/DC and DC/AC conversion, and galvanic isolation.
Provide arc-flash, ground-fault, and insulation monitoring; automatic fault localization and isolation.
Provide emergency jettisoning or safe-containment strategy (as certified/allowed).
Provide HV maintenance lockout/tagout, safe approach indications, and interlocked access.
Provide EMI/EMC control for HV cabling, shielding, bonding, zoning, and common-mode control.
3) Conventional Electrical Power (LV & MV)
Provide AC/DC generation (IDG/Starter-Generators/Alternators) and conversion (TRU/rectifiers/inverters).
Provide primary/secondary electrical distribution, protection, and load management.
Provide automatic source selection, paralleling, and bus tie logic.
Provide load shedding and electrical emergency configuration for critical loads (fly-by-wire, comms, instruments).
Provide power quality (frequency/voltage/THD) and phase synchronization.
Provide APU/eAPU (if applicable) start, power, and bleed functions.
Provide external power interface and protection.
Provide lightning/HIRF protection, bonding/grounding, and static dissipation.
4) Fuel & Thermal Management (for ICE/Generator)
Provide safe fuel storage, quantity measurement, and management (transfer, balance, jettison if equipped).
Provide engine/auxiliary power unit fuel feed with air/thermal/vapor control.
Provide inerting/venting and leak detection/isolation.
Provide fuel system icing protection and contamination management.
Provide integrated thermal management (heat sinks for motors/inverters/batteries via fuel/ECS/ram-air).
5) Flight Controls (Primary & Secondary)
Provide pitch/roll/yaw control authority and stability augmentation.
Provide fly-by-wire command/monitoring, actuator control, redundancy, and reversionary modes.
Provide high-lift system control (flaps/slats) with asymmetry detection and protection.
Provide speedbrake/spoiler control with load alleviation and gust suppression (if implemented).
Provide trim systems (pitch/roll/yaw) with runaway detection and cutout.
Provide envelope protection (AoA, g-limit, overspeed) and stall protection/shaker/pusher as applicable.
Provide jam/icing/fault accommodation, and manual reversion strategy as designed.
6) Autoflight & FMS
Provide LNAV/VNAV, RNP/LPV/ILS approach guidance and autothrust integration (including hybrid modes).
Provide autopilot/flight director modes, mode integrity/annunciation, and reversion.
Provide autothrottle/autothrust with energy-aware thrust requests to FADEC/EMDEC.
Provide performance management (takeoff/landing data, speeds, derates, noise abatement).
Provide trajectory management, time-of-arrival control, and 4D operations (as required).
7) Navigation, Communication, Surveillance
Provide air data/inertial reference, GPS/GNSS, radio nav (VOR/LOC/ILS/GBAS/SBAS).
Provide radios (VHF/HF), data link (ACARS/ATN, CPDLC), SATCOM (if fitted).
Provide ATC transponder/ADS-B/ELT, TCAS/ACAS X (as applicable).
Provide weather radar, Lightning/Stormscope, and terrain awareness (TAWS/EGPWS).
Provide integrated time source and clock synchronization.
8) Display Systems & Flight Deck Human Factors
Provide PFD/ND/MFD/EICAS/ECAM-like alerting and synoptics, with hybrid energy pages.
Provide alerting/annunciation with prioritization and inhibition logic.
Provide control interfaces (thrust levers, mode selectors, flap/gear handles, HV enable/standby).
Provide takeoff/landing configuration warnings (including energy mode constraints).
Provide checklists (normal/abnormal), electronic flight bag integration, and performance tools.
Provide external visual references (lighting, HUD/EVS if fitted).
9) Air Data, Sensors & Health Monitoring
Provide pitot/static, AoA, TAT, OAT, vibration, and structural load sensing.
Provide integrated data concentration, time-stamping, and parameter validation/voting.
Provide condition-based maintenance (CBM), HUMS for engines/motors/gearboxes/batteries.
Provide fault reporting, maintenance messages, and safe-download/recording (FDR/CVR/Crash-Protected Memory).
10) Hydraulic/Pneumatic/Actuation
Provide hydraulic power generation, distribution, and backup (including e-pumps).
Provide leak detection, isolation, and quantity/pressure/temp control.
Provide pneumatic/bleed management (from engines/APU/eAPU/e-compressors) with over-temp/over-press protection.
11) Landing Gear, Wheels & Brakes
Provide gear extension/retraction, door control, and uplock/downlock indication.
Provide alternate gear extension method.
Provide steering control (incl. e-taxi integration if motors in wheels or propulsors used).
Provide braking (normal/alternate/parking), anti-skid, autobrake, and brake temperature monitoring.
Provide tire pressure/temp monitoring and fuse-plug protection.
Provide takeoff/landing performance monitoring (rejected takeoff logic/alerts).
12) Environmental Control & Pressurization
Provide conditioned air, ventilation, and humidity control for cockpit/cabin/equipment bays.
Provide cabin pressurization, schedule control, and outflow/safety valve protection.
Provide avionics/electronics bay cooling, battery bay thermal control, and smoke removal.
Provide ozone/air quality/particulate monitoring as required.
13) Ice & Rain Protection
Provide wing, tail, engine inlet, probe, and windshield ice protection (bleed, electric, or hybrid).
Provide ice detection (primary/backup) and logic to prevent exceedance of ice accretion limits.
Provide rain removal (wipers/hydrophobic coatings/heaters).
14) Fire Protection
Provide detection and extinguishing for engines, APU/eAPU, nacelles, battery/inverter bays, cargo, lavatories.
Provide cabin/cockpit smoke detection and ventilation/smoke removal.
Provide fire handle logic, fuel/HV isolation, and extinguishing agent control.
15) Structures, Doors & Flight Controls Surfaces
Provide structural integrity, fatigue/damage tolerance, and load paths for hybrid system additions.
Provide doors/hatches/rams/ram-air inlets with position sensing and locks.
Provide emergency exits operation and indication.
16) Cabin Systems & Passenger Safety
Provide seats/restraints, oxygen systems (crew/pax), and PSU control.
Provide interior/cargo lighting and emergency lighting with independent power.
Provide water/waste systems with contamination protection.
Provide galley power management and high-load interlocks (coordinated with aircraft electrical).
Provide IFE/Connectivity isolation from safety-critical networks (cybersecurity segregation).
17) Safety, Security & Cybersecurity
Provide system safety monitoring, fault containment regions, and fail-safe/ fail-operational behaviors.
Provide physical security of flight deck and critical bays.
Provide data/network security, partitioning, secure boot, updates, and key management.
Provide robust maintenance access controls and logging.
18) Ground Operations & Servicing
Provide turnaround support: refuel/defuel, HV charging or fast-charge, GPU/air cart connections.
Provide safe ground handling with HV interlocks and live-voltage indicators.
Provide brake cooling/time-to-dispatch and battery cooldown/charge-to-dispatch functions.
Provide de-icing/anti-icing interfaces and procedures.
19) Abnormal & Emergency Functions
Provide emergency electrical configuration (batteries/eAPU/ram-air turbine if fitted).
Provide emergency descent and pressurization protection.
Provide thrust-asymmetry compensation and degraded control law.
Provide smoke/fumes removal procedures and dedicated ECS modes.
Provide ditching/overwater equipment and sealing (if applicable).
Provide post-event safeing: HV discharge, engine/motor spindown, and fuel/HV isolation.
How to use this list in your FHA (Failure Hazard Analysis) quickly
For each function, define failure conditions (lost, erroneous, partial/degraded, unintended, latent).
Map each to operational phase(s) and environment (icing, lightning, ETOPS, MEL).
Assess effects (crew workload, injuries, aircraft damage) → classify (No Safety Effect, Minor, Major, Hazardous, Catastrophic).
Derive safety objectives (DALs, integrity levels, detection coverage, independence, segregation).
Feed into PSSA (architectural mitigations), then SSA (evidence, FTA/FMEA, tests).
If you want, I can drop this into a structured spreadsheet (columns for Function, Failure Condition, Phase, Effect, Classification, Mitigations) and pre-fill the hybrid-specific items so you can start marking hazards right away.
Our Comments
While ChatGPT’s response isn’t necessarily awful, it clearly shows that it doesn’t know how to distinguish what is needed at an aircraft-level and what needs to be defined at a system-level. It is also confused about the differences between functions, requirements, and solutions.
An aircraft function describes the behaviour of the aircraft, whereas a system function describes the behaviour of the system that is required to support an aircraft function. For example, an aircraft function could be to provide a habitable environment for crew and passengers. The system function would be that the Environmental Control System (ECS) provides thermal airflow control for crew and passengers.
As for the differences between a requirement and solution, one of the system requirements for the ECS could be that it should modulate the cabin temperature to +/- 2 degrees Celsius from a crew-adjustable setpoint, between 18 and 30 degrees Celsius, under normal operation. The solution could be a thermal management controller that can read sensors and control actuators, and has some built-in test capabilities.
We asked ChatGPT for aircraft functions, but the output, “[p]rovide FADEC/EMDEC (engine + motor controller) command/monitoring, built-in test, and fault accommodation”, is a mix of system-level requirements and solutions rather than an aircraft function. “Provide de-icing/anti-icing interfaces and procedures” sounds like a random mix of words.
From the experiment above, we can see that AI has the ability to quickly consolidate a lot of information. ChatGPT was able to generate some form of an aircraft function skeleton. Yet, it is relatively weak in navigating subtlety and nuance. If the system is properly trained and used, it has the potential to create coarse content while permitting its human user to focus on subtle details that require critical thinking and judgment. It could also raise topics an engineer might have overlooked.
Bjorn.
Im really enjoying your work here
Thanks. Henry does the brunt of the writing. BTW, good commenting from you.
Totally agreed on this. It’s especially valuable to see what AI cranks out and have that contrasted with actual engineering tasks.
As I read through the AI response, the roots of its knowledge are clearly on display. It’s exceptional at collating materials that have been used in training. But it lacks any real understanding of what they mean or how to use them.
It feels like talking to someone who has been locked in a room and provided materials to study, but without a formal education. They’ve done a great job with what they’ve been asked to do, but without the context of experience and understanding, they could only really be used as an information resource.
I think that’s where we are at present. It’s still a tool and a resource, and it has the potential to reduce the time involved in certification to a limited degree.
But I suspect we will need to look at other methods to achieve real savings. And I’m sure Bjorn and his team have more in the pipeline. Really looking forward to future articles!!
Thanks again to everyone working on this series.
An area I have zero reference and finding it most interesting.
For replacing the regional jets the Maeve MJ500 makes a good achievable spec, not promising heaven and earth. But if the new Airbus get one pilot in cockpit (the other in computers on ground) certified for 200-250 pax it will be hard for Maeve with 2 pilots for max 100 pax.
I asked my (free) Gemini AI:
How can I ask an AI tool the question “I’m designing a Boeing 737 and Airbus A320 replacement aircraft. It’ll use a hybrid propulsion system. Would you generate a list of aircraft functions for the safety analysis?” in such a way that is provides a structured framework and includes sources and uses established aerospace safety analyses sources?
Using a compressed, simple prompt for a very big question and than concluding it is incomplete and inconsistent.. Often asking the question defines what you get. Specially with AI.
I use a phased approach, defining what I really want to know, what my purpose is, the target public, how I want the output to be structured and what requirements I have regarding sources to be used.
The more freedom you give AI, the more it takes, which is useful if you want creativity / new approaches. You have to set bounderies if your goal is different.
keesje,
Thank you for checking with Gemini. It offered something very generic with no concrete text or example to improve the prompt( assuming the last three paragraphs of the comment came from Gemini).
I think ChatGPT actually started the work reasonably well. The text before the functions list gave me the impression that it understood the context. It even understood the phased approach of safety analysis. The challenge is likely related to the training data. For this specific case, data providers probably didn’t tag certain text as aircraft-level or system-level in the training data. They probably also didn’t tag certain text as functions, requirements, solutions, or specification. Another hypothesis could be that the data source (e.g. the internet) has many poorly written functions lists, FHAs, or other safety analysis documents. These data were then used for training the AI.
It would be interesting to review the output of a properly trained AI when it becomes available.
Hi Henry, I wrote the last three paragraphs in the comment myself, why would I use AI?
The prompt that was proposed by Gemini to me:
”I am performing a Preliminary System Safety Analysis (PSSA) for a conceptual hybrid-electric aircraft designed as a replacement for the Boeing 737 and Airbus A320.
Would you generate a structured list of aircraft functions for this safety analysis?
Framework: Organize the functions hierarchically, from top-level aircraft functions down to specific system and subsystem functions.
Categories: Include categories for the following systems:
Propulsion System: Cover both the thermal engine and the electric motor components.
High-Voltage Electrical System: Include power generation, distribution, and energy storage (batteries).
Thermal Management System: Address cooling for high-voltage components and batteries.
Flight Controls: Consider the integration with the hybrid-electric system (e.g., thrust asymmetry).
Avionics and Flight Deck: Include the human-machine interface for managing the new propulsion system.
Analysis: For each function, provide:
The function’s primary purpose.
One or more plausible failure conditions.
The potential severity of the failure, using the following classifications: Catastrophic, Hazardous, Major, and Minor.
Sources: Ensure all information is consistent with established aerospace safety standards. Cite or reference relevant documents like SAE ARP 4754A, SAE ARP 4761, and applicable FAA and EASA regulations and advisory circulars for novel aircraft technologies.”
I use different AI tools, combining their different approaches. Sometimes I check the consistency/ compliance of output of one AI tool with another AI tool. I studied prompting cheat sheets widely available to enhance my AI skills and avoid irrelevant output.
keesje,
Thank you for the clarification. There was probably a misunderstanding. I was asking ChatGPT for a list of Aircraft Functions, nothing more. We still have to work on the AFHA and PASA. We also have to flow things down to the system-level. It wouldn’t be fair to ask AI to do a PSSA at this stage since there is no context.
Thanks for this very informative series.
In regards to Boeing and cert, the following is interesting and confusing but a lot in it. Down from the Podcast is a text breakout of what was said. You have to scroll a bit. I prefer the text to follow this kind of discussion.
https://aviationweek.com/podcasts/check-6/podcast-whats-behind-latest-boeing-777x-certification-delay
Thanks for posting this, it’s a really great interview. It confirms all the things we’ve been saying here about Boeing’s status right now.
It’s interesting that the FAA is developing the new certification procedures as they go, with Boeing’s help. It’s a collaborative process but it’s obviously much slower.
Also interesting that they both agree that it can’t remain in this development mode, and once the FAA and Boeing get the process down, it has to move quicker.
I think it will, but it’s difficult to predict when. Like I said none of us has insight. But you can see why Ortberg raised it as an issue.
And I agree with them that although the process can be made more efficient, it’s never going back to what it was before the MAX accidents.
Also really interesting about the NMA rumors. That would be a great thing but I’ll believe it when I see it.
You are welcome. I am hoping its taken as an adjunct to what Bjorn and Harry are doing.
Clearly its a complex undertaking and the amazing thing to me is it does not just bog down entirely (though 777X does appear to be slowed)
It’s reminiscent of the NTSB hearing testimony on the Alaska plug door blowout incident.
FAA had mandated a safety management system (SMS) for Boeing in 2015, but didn’t complete the rulemaking until 2024, after the incident. It’s just a really slow process if you don’t have a blueprint.
Seems like there is broad consensus that the FAA moves really slowly to develop procedures. But not consensus on how to fix it.
It’s clear that the solution from the 2010’s, to delegate more, traded loss of quality and safety for a more efficient process. Which is not an acceptable trade.
But delegating less makes the opposite trade, unless FAA staffing can be substantially increased.
Using a collaborative method as we see now, has been good for quality but hasn’t yielded results for efficiency.
Maybe we can ask Bjorn to do his next series on this. He might evaluate the regulator perspective, to complement the manufacturer perspective.
As well as compare and contrast the major regulators such as FAA and EASA, and how they may approach certification and rulemaking processes differently. That would be interesting to explore.
Agreed serious interest in Bjorns take.
I thought the original ODA was a good balance, with the Boeing paid inspectors reporting to the FAA not a Boeing manager. I would like to see a return to that.
Unfortunately while it helps one part of the process, the framework for that as stated is working slowly. I have always tried to come up with at least options for a problem, this one is so far outside anything I can offer I am just watching.
I suspect some of it involves FAA staffing. That is a two headed beast in Congress as well as pay/expertise.
Maybe another interesting one to ponder. Can you get the expertise that is equal to the designers and engineers?
I tend to think the innovative ones go private but do wonder if there are really good people that like to analysis and work that end in a quieter environment?
On my level I got to the point that at times I worked with the best engineers to resolve issues. Maybe the highlight was a really good one asked me what I thought on a problem?
I was somewhat shocked as I expected he would lead. I was sure I was missing something as it was a pretty stupid and obvious issue.
But I laid it out and he said the mirrored his thoughts.
Not too often I got off the leash, usually my inputs were kept inside the company and not allowed to back and forth with the contractors engineers.
It was an interesting balance but the core of it was not to get into fights with contractors but keep my side informed.
My stuff was on pretty basic level mechanically. Always was surprised they did not know it better.
Great laugh was “why do you not like this unit”? I had laid it out that there was some things very iffy (bearing between two fans) and odd working and we had experience with a different mfg that built a very good well thought out unit.
In the end we were stuck with it, if you can’t come up with something wrong (even though no one has one of those) then we are not changing it.
It worked sort of, but it was a constant headache.
Yes, it’s clear that ODA is essential, but how it’s done really matters.
Perhaps with a really solid ODA reporting hierarchy, the FAA would have more confidence in Boeing’s process.
It was interesting Guy Norris mentioned that a relatively few people at FAA are really holding Boeing’s nose to the grindstone on certification. I’ve heard similar things about Starliner at NASA, there were a few people that were adamantly insisting on more work, while the majority were in favor of certification.
But whatever, that’s a consequence of incidents that raise concerns. No incidents, no concerns.
I think contrary to looking at recent events, isolated issues, the background of the long 777-9 certification delays lays in the certification strategy approved by FAA in 2014. And congress driven FAA aircraft certification “streamlining” from 2012 (ARC).
2018 saw the FAA turnound, forced by foreign autorities.
https://leehamnews.com/2024/02/26/ultimately-congress-is-responsible-for-mess-at-faa-boeing-spirit-et-al/
Agreed, that could be counted among the root causes. The problem though, is that it doesn’t address the current issues, or provide any insight into fixing things as they now stand.
That is also the problem with many of the Boeing failure theories, such as the McD merger or the leadership that came out of it. As a historical matter, it’s worthy of pontification. But as a practical matter, it doesn’t address the current problems.
Agreed. Its the Canary in the Mine in how not to do it.
Go down that patch again and .,……..
But current part, a solution is needed – Boeing as they should has complained and now the ball is in FAA court and a tough ball to deal with.
COMMAC has something of the same issue. They are totally risk adverse (for good reason) but that also means their system is slowed down though its mfg not cert in their case.
No one other than China knows what Cert patch was followed.
Complained should have read COMPLIED.
A reprieve of sorts today for NASA. Secretary Duffy relented and instructed NASA to implement the Congressional budget for the next year, rather than the Presidential Budget Request.
Many groups inside and outside NASA have been fighting the Trump budget, which was devastating. It cut almost all aviation research. Now that can move forward again, including the Lockheed low-boom aircraft and the Boeing truss-braced wing.
Although Boeing halted the construction of the truss brace demonstrator, NASA and Boeing are still pursuing it in the wind tunnel and simulations.
An outstanding question now is whether NASA will pursue the DOGE staff reductions. It’s going to be tough going if the projects continue but the staff are RIF’d.
In other surprise news, NASA awarded a contract to Blue Origin to deliver the VIPER rover to the moon, which rescues that project as well. It had been in the dumpster after the launcher was delayed. Now it has a new ride to the moon.
I see AI Bots being of great use in design where you have a body of design work completed from previous aircraft. Then you can create a prompt such as……… using a 43Bxxxx skin panel assembly as a starting point, create a new assy using thermoplastic frames, stringers and skins. Incorporate paddle fitting terminal fitting provisions for join at the stringer locations. Provide stringer limber holes for condensate drainage as well as locations for EBond every x inches. Provide edge margin for skin join allowing for adhesive bonding and mechanical fastener locations to jig the parts during cure.
With thus as a starting point an experienced airplane guy engineer can train his bot to get the grunt work done, and if he creates 20 to 50 different bots to do different things, he can refine his prompts to get the AI refined to usable real fast ESPECIALLY if he works with a prompt guru to write the prompt in AI translating what he wants into what the AI can understand.
Everybody is doing this now by the way….
It gets even better as your bots know more. You could do things like say. Using this panel assembly, add 60 inches of +stations at station xxx and replicate window spacing. Viola, you have a stretch airplane. Having bots do the grunt work let’s a smaller more experienced engineering team move a lot faster that a lot of techs banging out Catia models of new stuff
You really have to understand how to use AI effectively instead of assuming an old engineer always know better. GIGO.
Realize that all documented engineering reports, study materials, investigation reports, trainingmaterials, published NASA, DLR, Tsagi research reports have been scanned, indexed and stored under that little IA icon on your computer screen.
It’s up to you to use it smartly & responsible, or not. Your colleagues will & the Chinese competitors before them, using all your lessons learned.
https://www.nasa.gov/artificial-intelligence/
https://www.klover.ai/boeing-ai-strategy-analysis-of-dominance-in-aerospace-defense/
https://www.aviacionline.com/how-the-faa-plans-to-use-artificial-intelligence-to-modernize-aviation
“Realize that all documented engineering reports, study materials, investigation reports, trainingmaterials, published NASA, DLR, Tsagi research reports have been scanned, indexed and stored under that little IA icon on your computer screen.”
And also scanned:
All the publications with incorrect data, reasoning and/or conclusions, including deliberately cooked info, fake news, propaganda, and puberal rubbish on Instagram and TikTok.
AI drinks from a tainted well of knowledge, and is too dumb to independently separate chaff from grain.
ChatGPT 5 has demonstrated this shortcoming dramatically.
Abalone AI is perfectly able to value sources and focus on some/ exclude some if you want so.. GIGO is still reality.
As Abalone said, the “AI” will become self-refential
over time, feeding on its own I-garbage and producing more of the same. It cannot “value sources”, because
it is not sentient, cannot think, or reason, or surmise.
When you check one “AI” against another “AI”, you’re falling into precisely this trap..
AI is just a dumb librarian with a vast memory.
Any “interpretation” it gives is essentially just a (weighted) average of (assumed) interpretations that it stumbled across when learning — including all the wrong intepretations in that mix.
It doesn’t have any intellect that can “value”, “focus” or “exclude” — it’s essentially only able to repeat, summarize (often poorly), and extrapolate and interpolate (often inaccurately).
It hallucinates, misintetprets verbs, can’t judge context, and has no innate value system.
And, as @Vincent pointed out, because it’s trawling its own output, feedback loops will only serve to increase the incidende of junk output.
That recent MIT report said that 95% of paying corporate users see no benefit from AI to efficiency.
A recent Chinese study indicated that it’s essentially only of practical use as a synthetic helpdesk agent.
A gimmic, at best.
Let’s combine Bots AI and thermoplastic composites in something simple.
Using the 737-8max as a starting point. Create a replacement floor beam at all constant section floor beam locations. Replace all aluminum floor beams having active PLM utilization in last 1 year as well as all floor beams with current or future effecticity. Identify on the face of the drawing in zone 4 sht 1 general notes ” this floor beam is identical to floor beam xx-xxxx for design purposes.” Additionally ADCN the existing metal floor beams with the corresponding composite floor beam part number in zone 4 of sheet 1 general notes.
This is a straight part change and is really easily doable right now.
My wife the inventory planner says they have been doing exactly this for a long time because they are already manipulating parts at the detail and assembly level. Design from the top down is not the way to get the best bang for your buck, per her… working from the part level up gets the airplane revised the easiest…… Her point is that ALL the existing MRP systems are actually low level complexity AI and because AI can read our minds now so to speak, shouldn’t surprise anyone its same oh same oh, just faster and a bit more intelligent
I think caution is needed though, on the presumptions that are made about what the AI is actually doing.
I remember Boeing instrumenting all their tooling, with the notion that inspection can be done in real time at assembly. On paper it looks great, you can capture non-conformance at the moment it’s created, and fix it then rather than having a separate inspection and rework. You don’t need all those inspectors and rework teams.
In practice it was quite different. It wasn’t assured that the instrumentation could capture every non-conformance, nor was it assured that Boeing could extract the non-conformance from the acquired data.
What happened in many cases, was that Boeing became aware of the non-conformance by physical inspection later down the line, or even by latent defect in service. Then they could go back and mine the acquired tooling data to uncover where and when it happened, which was still very useful in identifying other vulnerable airframes, by patterns in the data.
The lesson is to beware of certainty in complex systems where you lack awareness or understanding of what is really happening. By all means pursue those ideas, but realize the danger in relinquishing visibility and transparency.
This stuff happens even with the best of intentions. The FIU bridge collapsed, even though the design was from a premier bridge company, was reviewed by a independent consultant, by the state DoT, and by the university safety and compliance division. It was also built by a premier construction company with enormous experience.
Many eyes on the project, yet no one saw the flaw, except the workers who lifted it into place. They reported several times that the bridge had cracked when they lifted it, and the crack was growing with time.
Even then, the design flaw was missed because it was masked by an unconventional truss structure. Instead the crack was presumed to be a normal result of the bridge being as yet un-tensioned. So they tensioned it, at which point it collapsed. Some truss members were in fact massively overloaded, which tensioning made much worse.
The point being, that if you trained an AI to help design this bridge, it likely would make the same error, because there is nothing in the training materials to flag it. So you always must keep in mind that an AI is not actually reasoning these things out, it’s parroting what it’s been taught. It doesn’t really understand the engineering principles involved.
Rob.
Absolutely agree. That’s why an experienced engineer must run the bots. They are not so different than human tech designers running a Catia Station in that you hand them an instruction set and they report back to you with work to review. It’s the same iterative process except its potentially much faster and more consistent.
“So you always must keep in mind that an AI is not actually reasoning these things out, it’s parroting what it’s been taught.”
That is exactly what AI is outgrowing. You can ask AI to review the design and paperwork against the quality requirements. Look for possible lacunes, inconsistenties and ask to review past incident investigations of similar designs/projects. Analyse if they are of relevance for the current project, and why.
Ask to do a detailed SWOT / risk analyses of all available documentation / reporting data, against relevant quality requirements / goals. In hours iso weeks. And if you feel the result is not complete, focussed enough, you have progressing insights, adjust the request & run it again.
It’s not the question if it’s done like this, you’re behind already if you didn’t.
One chief engineer told me “If you count of friction for the structure to work you will not have it, if you count on little friction you will sometimes have lots of it”.
He was in aerospace not bridge design:
The engineering design error that directly led to the collapse was identified by the National Transportation Safety Board (NTSB) as a miscalculation of resistance to sliding of the connection between the walkway surface and the truss that held it up
This is a good overview of the stashed 737s and 787s. It lays out in some detail the maint and the getting back into service that is required.
https://www.youtube.com/watch?v=aWV4RdkgxsY
One of the less covered aspects is having to dissemble 787s to get the shim work right. While not all 787s in service had the dual issue (Shim wrong and fuselage out of tolerance) none of the one in Boeing’s hands could be delivered until both were corrected.
Just keep software updated on both types would be time consuming.
Insufficient FAA Certification personnel to perform Certification Workload.
Ok if we design more product on shorter timelines we are proceeding forward with the expectation that the FAA will be able to meet our timelines for Certification Issuance. Apparently they know all about the understaffed AIR organisation’s that are seriously stretching project timelines.
As a user of the system having run my own packages through the system, I can tell you that today it was very difficult to get the right people to review my latest submission. So much so that I shelved the project. The GAA apparently recognizes thia issue and has created a look forward document to guide their future staffing thoughts. Perhaps this is what is behind Ortbergs frustration with sequential TIA and the implication that it is adding to delays. The reality of the aerospace industry in America is that the product innovators generally do not hold FAA job titles and the FAA engineers are always playing a bit of catch up.
Here’s the FAAs answer to how they get the Certification staffing back up to where needed to service the users.
https://www.faa.gov/about/office_org/headquarters_offices/afn/offices/finance/offices/office-financial-labor-analysis/plans/avs-workforce.pdf
Boy I wish ABALONE would rebut the FAAs stance that they do not have enough Safety Personnel to do the Certification Workload despite his assurances to us that they are properly staffed…..
I already did — twice.
And I didn’t give an “assurance” — I made an assertion…just like you’re doing.
Did you actually read the link you posted?
It predominantly talks about replacing run-off…with relatively little net increase.
Also: have you considered that having large “babysitting” teams at a major OEM is not part of the normal “certification workload”, and is a drag on staffing?
That’s essentially a Boeing screw-up — not an FAA one.
It seems the chinese are in the final stages of testing and certification of magnetic catapults in combination with various new aircraft types.
https://x.com/OedoSoldier/status/1970067493529301444