It’s hard to imagine a worse way to be awoken on a Saturday morning in paradise than with a blaring klaxon accompanying a government alert about an inbound ballistic missile attack. But that’s exactly what happened to more than 1.5 million people in Hawaii this morning.“BALLISTIC MISSILE THREAT INBOUND TO HAWAII,” the emergency alert read, in all-caps, on smartphones. “SEEK IMMEDIATE SHELTER. THIS IS NOT A DRILL.” Hawaii residents tuned-in to television or radio heard an even more threatening message, made worse by its monotone, computer-synthesized delivery. “The U.S. Pacific Command has detected a missile threat to Hawaii. A missile may impact on land or sea within minutes. THIS IS NOT A DRILL.”

Except, it was indeed a drill—there was no missile threat, and the alert had been sent in error during what the Hawaii Emergency Management Agency called a “regular system drill.” Easily screencapped, the mobile alerts spread like wildfire within minutes. On Twitter, Tulsi Gabbard, who represents Hawaii’s second district in the U.S. House of Representatives, sent her own all-caps alert, desperately trying to assuage citizens and visitors that the message was erroneous: “HAWAII – THIS IS A FALSE ALARM. THERE IS NO INCOMING MISSILE TO HAWAII.”

War and politics notwithstanding, what makes such a false alert possible in the first place? Most Americans don’t know how emergency alerts work. Both the infrastructure for sending these notifications and the media ecosystem into which they arrive have changed substantially since the Cold War, when the shadow of nuclear annihilation last felt near.

* * *

In 1997, the Emergency Alert System (EAS) came online, replacing the Emergency Broadcast System (EBS), which had been in place since 1963. To those old enough to remember broadcast television and radio, the EBS was a ubiquitous part of media life during the Cold War, which was also the era of television as a predominant information-delivery mechanism. Like every system, the EBS issued regular tests, and every 20th-century American citizen associated the television or radio voiceover “This is a test of the Emergency Broadcast System” with the ubiquitous shadow of global catastrophe.

Both EBS and EAS were designed to allow the president of the United States a channel to communicate quickly to the American public in the event of national crisis. That mostly meant war in the early days of EBS. Later, the system was used to provide notice about other sorts of emergencies, including natural disasters, severe weather, and other local civic emergencies. EAS formalized that function, which had become the primary purpose of EBS before its retirement. EAS also addressed the profusion of broadcast channels present in the mid-1990s, as compared with the 1960s: not just AM and FM radio, and broadcast television, but also cable, fiber, digital, and satellite television, satellite radio, and more.

In 2006, after criticism surrounding government preparedness and response during Hurricane Katrina, President George W. Bush established a new program, the Integrated Public Alert and Warning System (IPAWS). IPAWS integrated EAS and the other government warning systems, including National Warning System (NAWAS), an automated telephone warning system; the Commercial Mobile Alert System (CMAS), a warning system for mobile devices; and the National Weather Service’s Weather Radio system.IPAWS remains in place today, and it is the service used to send the erroneous missile alert to everyone on the islands of Hawaii today. The whole system is managed by FEMA, which authorizes individual local agencies to send emergency messages. In Hawaii, there is only one authorized agency, the Hawaii Emergency Management Agency.

* * *

With such sophisticated and time-tested systems behind the scenes, how could that agency make a mistake like this one? The simple answer is human error, which is how the Hawaii Emergency Management Agency spokesperson Richard Rapoza characterized the incident in an interview with my colleague Adrienne LaFrance. That explanation serves up cold comfort to the Hawaii residents who thought they faced possible obliteration this morning. “My 10-year-old said, ‘I thought we were going to die,’” tweeted Chris Gaither, who is on vacation in Hawaii.

But beyond a bad look, human error is also not a terribly informative explanation for the spike in anxiety the message created. That’s largely because emergency notifications have become so efficient. Too efficient, maybe.

In 2013, FEMA replaced the Commercial Mobile Alert System with Wireless Emergency Alerts (WEAs). These are the messages you receive on your smartphones today. They include AMBER alert notices for child abductions, government emergencies including weather and, God help us, ballistic missile attacks, and direct messages from the president. If you look carefully at your smartphone’s notification settings, you’ll see that you can disable the first two kinds of alerts, but not the third—which has never been used since the conception of all these systems in the mid-20th century.Consider Gaither’s account of how his family learned about the fictional missile attack. “We’re an iPhone family and all our devices shrieked Amber-style alerts at the same time.”

Most people have had just this experience. IPAWS messages include an allocation for WEAs, which are sent directly to participating wireless carriers. Those messages are pushed to mobile devices active on cell towers in the area where the alert is defined—that’s why you can receive an emergency alert while traveling. Like the old EBS, an ear-piercing sound precedes a WEA, meant to grab the attention of its recipient.

People have their smartphones with them all the time. Even in bed, or nearby. That makes WEAs an effective way to reach people immediately and directly in the event of an emergency. But it also means that there is no slack in the system. The old EBS notices, or any broadcast message sent via EAS, only reached citizens who were watching television or listening to radio at the time it was sent. The system would repeat the messages occasionally, but there didn’t used to be a way to reach so many people all at once. That can be good if the information is sound, or terrible if it is not, as was the case in Hawaii today.

But perhaps the information in a WEA can never be sound, because it cannot be self-contained. A WEA must be 90 characters or less; there’s no room for elaboration or further instruction. The EAS message the Hawaii Emergency Management Agency sent to television and radio broadcasts included instructions for taking shelter, at least—the wireless notice did little more than suggest unvarnished panic. That’s all it can do.Back in the days of broadcast, emergency-notification messages often instructed affected citizens to tune in to local media for further information and instruction. An emergency notice is mostly a jolt to take action—but what action? And how? And when? And why? In the 1980s or even the 1990s, that would have involved tuning in to radio or television. Even on the morning of 9/11, the internet was useless—servers overtaxed, just as the Hawaii Emergency Management Agency’s were today.

But there is hardly any local media to tune into, anymore. In some cases it’s been shrunk or even shuttered, as global tech companies have imposed on its viability. In other cases, people have chosen to “cut the cord,” severing their ties to both broadcast and cable in favor of globalized, streamed entertainment.

Where local media remains, it’s also global. Twitter and Facebook, where word of the false alarm spread first and most rapidly, have hardly established themselves as reliable platforms for news in the last year. And besides that, the millions (or billions) of people who saw the false alarm on social media probably got in the way of the much smaller community of people on the ground in Hawaii who really needed less noise rather than more as they evaluated the situation, and then as they recovered from the mental anguish it had incited.

* * *

Even so, Americans far beyond the Central Pacific, already reeling from geopolitical anxieties of all stripes, were calling for the head of whoever made the error. “We needed a cancellation procedure,” Rapoza, the Hawaii Emergency Management Agency rep, told LaFrance, who contributed reporting to this story. It does appear that the agency might have attempted to cancel the IPAWS alert five minutes after sending the erroneous one. But even if that cancellation might have ceased further delivery of a damaging message, it wouldn’t have issued a reversal.

That’s because WEA “worked the way it was supposed to,” as Rapoza put it. These aren’t like text messages, where a sender can dash off a quick sorry my badif they mistype. IPAWS notices have a specific format, which must be composed formally and in advance. Audio files for broadcast notices must be recorded or generated and uploaded. Often, this has to be done by special software on special equipment.

According to Rapoza, the agency was undergoing a shift change. “During shift change they do a drill and somebody clicked the wrong thing on a computer,” he said. To send a meaningful all-clear notice, which appeared about a half hour later (“There is no missile threat or danger to the State of Hawaii. Repeat. False Alarm.”) required properly composing and offsending a new IPAWS civil emergency message.

So how could this happen? For many reasons. In part, it’s because emergency systems have long decoupled from the threat of war in general, let alone global thermonuclear war specifically, even as they’ve become incredibly robust at sending peacetime messages. In part, it’s because media have splintered and fragmented, making it hard to get detailed official messages to everyone. In part, though, it’s because media have consolidated in devices everyone holds in their hands and pockets, but which work best with small quantities of narrow-bandwidth information. And, in part, it’s because someone pressed the wrong button on a computer, which then did exactly what it was programmed to do.Traditional automatic speech recognition (ASR) systems, used for a variety of voice search applications at Google, are comprised of an acoustic model (AM), a pronunciation model (PM) and a language model (LM), all of which are independently trained, and often manually designed, on different datasets [1]. AMs take acoustic features and predict a set of subword units, typically context-dependent or context-independent phonemes. Next, a hand-designed lexicon (the PM) maps a sequence of phonemes produced by the acoustic model to words. Finally, the LM assigns probabilities to word sequences. Training independent components creates added complexities and is suboptimal compared to training all components jointly. Over the last several years, there has been a growing popularity in developing end-to-end systems, which attempt to learn these separate components jointly as a single system. While these end-to-end models have shown promising results in the literature [2, 3], it is not yet clear if such approaches can improve on current state-of-the-art conventional systems.

Today we are excited to share “State-of-the-art Speech Recognition With Sequence-to-Sequence Models [4],” which describes a new end-to-end model that surpasses the performance of a conventional production system [1]. We show that our end-to-end system achieves a word error rate (WER) of 5.6%, which corresponds to a 16% relative improvement over a strong conventional system which achieves a 6.7% WER. Additionally, the end-to-end model used to output the initial word hypothesis, before any hypothesis rescoring, is 18 times smaller than the conventional model, as it contains no separate LM and PM.

Our system builds on the Listen-Attend-Spell (LAS) end-to-end architecture, first presented in [2]. The LAS architecture consists of 3 components. The listener encoder component, which is similar to a standard AM, takes the a time-frequency representation of the input speech signal, x, and uses a set of neural network layers to map the input to a higher-level feature representation, henc. The output of the encoder is passed to an attender, which uses henc to learn an alignment between input features x and predicted subword units {yn, … y0}, where each subword is typically a grapheme or wordpiece. Finally, the output of the attention module is passed to the speller (i.e., decoder), similar to an LM, that produces a probability distribution over a set of hypothesized words.

Components of the LAS End-to-End Model.

All components of the LAS model are trained jointly as a single end-to-end neural network, instead of as separate modules like conventional systems, making it much simpler.
Additionally, because the LAS model is fully neural, there is no need for external, manually designed components such as finite state transducers, a lexicon, or text normalization modules. Finally, unlike conventional models, training end-to-end models does not require bootstrapping from decision trees or time alignments generated from a separate system, and can be trained given pairs of text transcripts and the corresponding acoustics.

In [4], we introduce a variety of novel structural improvements, including improving the attention vectors passed to the decoder and training with longer subword units (i.e., wordpieces). In addition, we also introduce numerous optimization improvements for training, including the use of minimum word error rate training [5]. These structural and optimization improvements are what accounts for obtaining the 16% relative improvement over the conventional model.

Another exciting potential application for this research is multi-dialect and multi-lingual systems, where the simplicity of optimizing a single neural network makes such a model very attractive. Here data for all dialects/languages can be combined to train one network, without the need for a separate AM, PM and LM for each dialect/language. We find that these models work well on 7 english dialects [6] and 9 Indian languages [7], while outperforming a model trained separately on each individual language/dialect.

While we are excited by our results, our work is not done. Currently, these models cannot process speech in real time [8, 9, 10], which is a strong requirement for latency-sensitive applications such as voice search. In addition, these models still compare negatively to production when evaluated on live production data. Furthermore, our end-to-end model is learned on 22 million audio-text pair utterances compared to a conventional system that is typically trained on significantly larger corpora. In addition, our proposed model is not able to learn proper spellings for rarely used words such as proper nouns, which is normally performed with a hand-designed PM. Our ongoing efforts are focused now on addressing these challenges.

Acknowledgements
This work was done as a strong collaborative effort between Google Brain and Speech teams. Contributors include Tara Sainath, Rohit Prabhavalkar, Bo Li, Kanishka Rao, Shankar Kumar, Shubham Toshniwal, Michiel Bacchiani and Johan Schalkwyk from the Speech team; as well as Yonghui Wu, Patrick Nguyen, Zhifeng Chen, Chung-cheng Chiu, Anjuli Kannan, Ron Weiss, Navdeep Jaitly, William Chan, Yu Zhang and Jan Chorowski from the Google Brain team. The work is described in more detail in papers [4-12].

References
[1] G. Pundak and T. N. Sainath, “Lower Frame Rate Neural Network Acoustic Models,” in Proc. Interspeech, 2016.

[2] W. Chan, N. Jaitly, Q. V. Le, and O. Vinyals, “Listen, attend and spell,” CoRR, vol. abs/1508.01211, 2015

[3] R. Prabhavalkar, K. Rao, T. N. Sainath, B. Li, L. Johnson, and N. Jaitly, “A Comparison of Sequence-to-sequence Models for Speech Recognition,” in Proc. Interspeech, 2017.

[4] C.C. Chiu, T.N. Sainath, Y. Wu, R. Prabhavalkar, P. Nguyen, Z. Chen, A. Kannan, R.J. Weiss, K. Rao, K. Gonina, N. Jaitly, B. Li, J. Chorowski and M. Bacchiani, “State-of-the-art Speech Recognition With Sequence-to-Sequence Models,” submitted to ICASSP 2018.

[5] R. Prabhavalkar, T.N. Sainath, Y. Wu, P. Nguyen, Z. Chen, C.C. Chiu and A. Kannan, “Minimum Word Error Rate Training for Attention-based Sequence-to-Sequence Models,” submitted to ICASSP 2018.

[6] B. Li, T.N. Sainath, K. Sim, M. Bacchiani, E. Weinstein, P. Nguyen, Z. Chen, Y. Wu and K. Rao, “Multi-Dialect Speech Recognition With a Single Sequence-to-Sequence Model” submitted to ICASSP 2018.

[7] S. Toshniwal, T.N. Sainath, R.J. Weiss, B. Li, P. Moreno, E. Weinstein and K. Rao, “End-to-End Multilingual Speech Recognition using Encoder-Decoder Models”, submitted to ICASSP 2018.

[8] T.N. Sainath, C.C. Chiu, R. Prabhavalkar, A. Kannan, Y. Wu, P. Nguyen and Z. Chen, “Improving the Performance of Online Neural Transducer Models”, submitted to ICASSP 2018.

[9] C.C. Chiu* and C. Raffel*, “Monotonic Chunkwise Attention,” submitted to ICLR 2018.

[10] D. Lawson*, C.C. Chiu*, G. Tucker*, C. Raffel, K. Swersky, N. Jaitly. “Learning Hard Alignments with Variational Inference”, submitted to ICASSP 2018.

[11] T.N. Sainath, R. Prabhavalkar, S. Kumar, S. Lee, A. Kannan, D. Rybach, V. Schogol, P. Nguyen, B. Li, Y. Wu, Z. Chen and C.C. Chiu, “No Need for a Lexicon? Evaluating the Value of the Pronunciation Lexica in End-to-End Models,” submitted to ICASSP 2018.

[12] A. Kannan, Y. Wu, P. Nguyen, T.N. Sainath, Z. Chen and R. Prabhavalkar. “An Analysis of Incorporating an External Language Model into a Sequence-to-Sequence Model,

I made some pottery from the clay in the new area to see how well it performed. A large bank of clay was exposed by the side of the creek. I dug it out using a digging stick and took it back to the hut. Small sticks and stones were picked out of the clay and the whole mass was mixed to make sure there were no dry lumps. When this was done the clay was then left next to the fire to dry slightly so that it became a stiff workable material to form pots from. No further processing was done to the clay.

I formed small pinch pots from the clay by taking balls of it and pinching out the shape of the pots. Small cracks that formed while shaping were simply mended by wetting and smoothing over. Several pots were made this way. They were then left to dry completely next to the fire until they were completely dry.

To fire the pot, it was placed upside down in the hot coals and covered with sticks in a tipi fashion. The wood both acts as fuel and protects the pot from sudden changes in temperatures such as those caused by sudden winds. When the fire was burning well, I increased the temperature of the fire by fanning it with a fan palm frond. The pot glowed red hot amongst the coals and so was fired to a sufficient temperature. After waiting overnight, the pot was retrieved from the ashes and struck with a stick. The pot gave a clear ringing sound indicating it was strong and had no cracks (hollow sounds indicate the opposite).  Now I had a small bowl to carry water in.

A larger pot was then made from the same clay. This time the walls of the pot were built up using the coil technique where long rolls of clay were rolled and then squashed onto previous layers. The last layer was pinched outwards to form a pot lip. A lid was made for the pot by making a flat disk of clay with a small handle for lifting. When dried the pot was then fired as before but in a larger pit outside the hut. Again, the pot was covered with wood protecting it from sudden breezes that might cool or heat the pot suddenly, possibly causing cracks. The firing went well and the pot sounded strong when struck.

The pot was then placed on 3 rocks and a fire lit underneath. It took close to 30 minutes to boil this way with lots of sticks. But it did eventually come to the boil. I then made a stove inside the hut. The fire pit was dug and extended into a trench, sticks laid over the entrance and mud mixed from the excavated dirt was then used to form the walls of the stove over the trench. The stove was about 30 cm internal diameter but came in to about 20 cm. Three raised lumps were made on the top of the stove to hold the pot above. Then the stove was fired. Note that wood can be placed over the entrance of the stove at ground level and lit in a hob firebox like configuration. The flames then get sucked down and then up into the stove. I show this because it’s an easy way to manage the fire without making it too big which might burn the thatch.

When the pot is on the stove, it’s easier just to put sticks straight into the top of the stove between its open top and the sides of the pot. If over stacked with wood, wood gas is produced burning in a second fireball above the stove. It’s best just to keep the flames big enough to surround the pot (to reduce fire hazards). The pot was quicker to come to the boil then over a three stone fire.

The clay here in the new place is good, it didn’t take me long to make pottery here. Notably this clay doesn’t seem to need grog or temper added to it to prevent it from cracking. I think this is due to tiny specs of mica that weren’t present in the clay from my old area. The clay seems stronger and there also seems to be much more of it everywhere. The pot boiled after a while of tending, in future I’ll probably make thinner walled pots so that they boil quicker. The stove was useful for boiling the pot. It also seems to reduce the amount of smoke in the hut and increase the life of the coals in the base so that the fire could be re stoked at a later time.

New area Starting from scratch

I bought a new property to shoot primitive technology videos on. The new area is dense tropical rainforest with a permanent creek. Starting completely from scratch, my first project was to build a simple dome hut and make a fire. First, I took some wood, Abroma mollis, for fire sticks. I knapped a small stone blade and used it to strip the fire sicks. Palm fibre was then taken for the tinder. The fire stick kit was then placed under a palm leaf to keep it out of the rain.

Next, a stone from the creek was fashioned into a simple hand axe. This was used to cut a staff that was used to clear a path to the hut location. The location for the hut was a clearing densely crowded by native raspberry. This was then cleared using the staff and a small 2.5 m circle was levelled ready for building.

Eight 2.75 m long saplings were cut using the hand axe and brought to the site. Eight holes about 25 cm deep were hammered into the ground in a circle 2.5 m in diameter and the saplings were then planted in. The tops were brought together at the top and tied with vine. A door lintel stick was lashed to the front about 75 cm off the ground giving a low door way.

A stone flake was used to cut about 600 palm fronds. These were split and lashed horizontally to the frame creating a thatched dome. Mosquitoes are a real problem here so a fire was lit. The fire sticks from before had a hole carved in the base boards and had a notch carved to let the powder pour out.

The spindle was twirled in the socket and smoking powder poured out producing a hot coal. This then ignited the palm fibre tinder. The fire was transferred to the hut and a small hearth was made of stones. The fire makes a big difference in the number of mosquitoes which seem unable to tolerate the smoke. The dome was completed up to the top and a small cap was made from lawyer cane and fronds to place on the top to keep rain out. When not in use the cap can be removed to let in more light like a sky light.

Finally wood was cut for a bed. This consisted of wooden stakes hammered into the ground at the back of the hut behind the fire pit. Part of the bed frame is attached to the sapling uprights that form the dome. This works ok without the frame shaking too much due to the low attachment point of the bed. Wooden boards were then placed on this and were covered with palm fibre for bedding. Firewood is stored just inside the entrance on the left side of the door looking in. The bed sits behind the fire pit so smoke and flames deter insects or large animals reaching the occupant. Fire sticks and tools are kept just inside the right side of the entrance.

The small hut is simple to build and creates a small, dry shelter for camping and storing tools. Though it is dark, the cap can be removed in fine weather to provide a fairly well-lit workspace protected from annoying insects. This new area has good stone, clay and materials lending themselves to elaborate shelters. A permanent creek runs through it. Mosquitoes are abundant here though and will be an issue. The Cassowary, a large, horned, flightless bird lives in this forest. It’s the most dangerous bird in the world, but generally only attacks when threatened.

 

 

 

Natural Draft Furnace

I built a natural draft furnace to test ideas about how hot a furnace could get without the use of bellows. Natural draft is the flow of air through a furnace due to rising hot air. The hot gasses in the fuel bed are more buoyant than the cold air outside the furnace causing them to rise. Fresh combustion air then enters the base of the furnace to replace the rising combustion gasses, keeping the fuel bed burning. This effect increases with: 1. the average temperature of the fuel bed relative to the outside air and 2. The height of the furnace. Two other important factors are the size of the tuyere (air entry pipe) and lump size of the fuel bed as these effect the resistance to airflow through the furnace. The furnace was tested with wood fuel and some ore was melted but produced no iron. High temperature were indeed produced (probably about 1200 c). These types of furnaces were once used for smelting copper and iron ores in around the world in ancient times, usually using charcoal as a fuel and in some cases wood too.

I designed the furnace using a formula from the book “The mastery and uses of fire in antiquity” by J.E. Rehder. It was designed to have a space velocity (air speed within the furnace) of 6 m per minute which is recommended for iron smelting. The furnace was 175 cm in total height but with a height of only 150 cm above the tuyere. The height between the air entry and the top of the furnace is what determines the strength of the draft, the space beneath the air entry is not included in the formula. The internal furnace diameter was 25 cm. The walls were about 12.5 cm thick at the base but got thinner with height. The tuyere (air entry pipe) was 7.5 cm internal diameter and about 20 cm long. The tuyere was placed into an opening in the base of the furnace and sealed with mud. The whole thing took about a week to make due to the slow drying time that was assisted by keeping a fire burning in side it. The furnace was designed to use charcoal (which in this case should be 2.5 cm diameter lumps) but I used wood to test it instead as it was easier to acquire. To test its melting ability, bog ore was found further down the creek and roasted. The roasted ore was then crushed and stored in a pot.

The furnace was filled with wood and lit from the top. The fire burnt down the furnace producing charcoal. On reaching the tuyere the fire then started burning the charcoal. Wood was also continually added from the top along with a few small handfuls of the roasted bog ore (not shown in the video). The temperature of hot objects can be visually estimated from their incandescence.  After about an hour, the light coming out of the tuyere was high yellow to white hot indicating a temperature of at about 1200 c. Colour temperature charts vary but white hot is usually given to be at least 1200 c, examples of these charts can be found on the internet for reference. It was uncomfortable to stare into the tuyere and doing so left an after image when looking away, indicating the strength of its brightness. After about an hour and a half the furnace was left to burn out. When opened the next day the tuyere was covered in slag with bits of slag found on the furnace floor also.

This experiment shows that high temperatures can be achieved without the use of bellows or charcoal, which might significantly reduce labour in the production of iron. The furnace was technically easy to build as it was a simple vertical cylinder. When running, the wood added to the top of the furnace converts to charcoal in the upper part of the stack and is consumed in the lower part. The ore I used was new to me, normally I use iron bacteria as an ore. This new ore produced no metallic iron so I’m inclined to use iron bacteria in future. Natural draft furnaces were once used to smelt copper and iron ores in the past, usually with charcoal fuel and less frequently with wood. The main benefit of these furnaces seems to have been the reduction in labour they provide and simplified infrastructure (fewer workers and no bellows required during operation).

Mud bricks

I made a brick mold that makes bricks 25 x 12.5 x 7.5 cm from wood. A log was split and mortise and tenon joints were carved using a stone chisel and sharp rocks. The mold was lashed together with cane to prevent it from coming apart when used.

Next, I made a mixture of mud and palm fiber to make the bricks. This was then placed into the mold to be shaped and taken to a drying area. 140 bricks were made.
When dry, the bricks were then assembled into a kiln. 32 roof tiles were then made of mud and fired in the kiln. It only took 3 hours to fire the tiles sufficiently. The mud bricks and tiles were a bit weaker than objects made from my regular clay source because of the silt, sand and gravel content of the soil. Because of this, I will look at refining mud into clay in future projects instead of just using mud.

Interestingly, the kiln got hot enough so that iron oxide containing stones began to melt out of the tiles. This is not metallic iron, but only slag (iron oxide and silica) and the temperature was probably not very high, but only enough to slowly melt or soften the stones when heated for 3 hours.

The kiln performed as well as the monolithic ones I’ve built in the past and has a good volume. It can also be taken down and transported to other areas. But the bricks are very brittle and next time I’d use better clay devoid of sand/silt, and use grog instead of temper made of plant fiber which burns out in firing. The mold works satisfactorily and I aim to make better quality bricks for use in furnaces and buildings in future.

Simplified blower and furnace experiments

Blower description

The purpose of this project was to test a simplified blower design connected to a furnace. I purposely did this to show that people in most natural environments should be able to replicate this design without difficulty. This blower differed from the previous one in several ways to simplify the construction method.

Firstly, the impellor was simply a stick as a rotor with a 40 cm wide rectangle of bark tied into a split in its end with a bark fibre cordage. A stone with a pit carved into it acted as a socket for the lower half of the rotor to spin in. If spun in the dirt the rotor can drill down and the position of the impellor can reach ground level causing the blades to bump into rocks and dirt. Later, I plastered the stone socket into the ground with mud to hold it securely in position (not shown in the video, just be aware of this solution if the socket shifts around too much).

Secondly, the housing for the blower was made in situ of ordinary mud (dirt and water on site). It was a bit more than 40 cm in internal diameter. The walls of the housing were solid mud and the roof was made of sticks covered with mud. An opening more than half the length of the impellor was left in the roof to remove the impellor for maintenance and to admit air into the blower during operation. In use, the portion of this opening near the front of the blower was covered with a tile. If left opened the blower still worked but covering it improved performance by preventing air escaping near the front. In places where water is not available, a housing shaped pit covered with sticks and dirt might work instead.

Finally, a simple length of cordage was used to drive the rotation of the impellor. This cord was placed in a notch carved into the top of the impellor rotor. The cord was wrapped around the rotor about 2.5 times. During operation the cords were pulled outwards causing the rotation. When fully unwound, the momentum of the impellor then wrapped the cord back around in the other direction. Then the cords were pulled outwards again causing the impellor to spin in the other direction. Note that this is a centrifugal blower with a symmetrical housing, therefore it doesn’t matter whether the fan spins one way or the other (clockwise or anti clockwise), the blower will always suck air in to its open top and force it into the furnace.

This design is easy to make and use. It can be made with minimal materials by unskilled people. The impellor design is simple yet effective. A stick, some bark and lashing of some sort should be available in most areas. The housing being made from mud, is easily sourced also. For the drive mechanism, I chose this method because the first blower I built had too many parts. There was a frame made of wood and vine to hold the rotor in place which kept causing issues with the rotor seizing or jumping out of the socket. Also, the bow that was used to drive the rotor added unnecessary complexity. In the new design, the simple cord in the notch of the rotor did away with the frame and the bow of the old design and the associated difficulties.

Furnace experiments

The blower was used to power a furnace attached to the front of it. Note that with minimal materials, the blower could simply force air into a hole in the base of the furnace and work satisfactorily. But I wanted to test a different configuration so I used clay grate from a previous kiln I made. Fuel in the form of wood and charcoal was used in this furnace by being placed on top of the grate instead of under it. During operation, the blower forced air up through the grate into the burning fuel bed increasing the rate of heat production relative to the use of natural draft (convection) alone.

I made 3 pots and fired them with charcoal. The first pot was painted with iron bacteria (iron oxide being the active ingredient). When fired, the oxide melted slightly showing minimal glazing. The clay became quite hard, possibly stoneware. The second pot was painted with wood ash and placed on a three sided clay plinth to hold the pot in the position of highest temperature in the fuel bed. The pot softened and sagged apart catastrophically. But the ash glaze gave a dark green smooth finish (difficult to see in the video). Finally, a pot was place upside down on the grate and a cylindrical brick made of iron bacteria, charcoal powder and wood ash was put on top of this. The brick melted over the pot, covering it in a viscous blob of slag rather than a thin glaze. On inspection, the slag had 1 mm sized spheres of metallic iron in it. Some of these were picked out and stored in a pot. The reason for these experiments like these to gain knowledge that might be of practical use in future projects that have not yet been determined.

Sandals

I made a pair of sandals from loya cane. Walking bare footed in the bush generally doesn’t cause problems for my feet. But when repetitively carrying loads of various materials the soles of the feet become cracked and split. So I made some basic footwear for the purpose of working on rough surfaces.

I cut some cane and measured out a length 6 times the length of the foot (about 1.5 m), folded it into loops and wove more cane between the loops to form the sole, adding new cane as needed. Next, I made bark fiber cordage and threaded it through the sandal to keep it on. The pair took about 1 hour to make (longer due to setting up the camera).

The sandals do protect from the ground, preventing the feet from cracking. I personally don’t like wearing footwear in the forest as bare feet give better grip, especially on inclines. But for heavy work or when my feet are injured I’ll wear these. These sandals are so quick to make that I’ve already got 2 pairs. The material used to make them (loya cane) is everywhere here but pretty much any rope like material will do. Bark fiber rope, grass, vine, flexible roots etc. will all make usable alternative materials.

Reusable Charcoal Mound

Charcoal is a valuable fuel that reaches a higher temperature than the very wood it’s made from. I’ve made some before, but with supplies running low due to furnace experiments, I decided to make another large batch of charcoal in a mound. I stacked the wood into a roughly conical shape (about 1 m wide and 75 cm high) and then built a thick wall of mud around the heap (this took 6 hours). Eight air entries were made in the base of the mound and one air exit hole was left at the top of the mound to allow the volatile components of the wood to escape while creating a natural draft to keep everything burning.

The mound was lit and the flame burned backwards down the heap in the opposite direction to the draft. This protects the coal made above the level of the fire from burning as carbon dioxide rushes past instead of oxygen, preventing combustion of charcoal. Each air entry was sealed only when fire became visible through them. This is an easy way to tell when to close them up, i.e. when the fire had burned down all of the wood in the heap. When the last air entry was closed, the air exit at the top of the mound was sealed, 5 hours after starting. The next day when cool, a large arched opening was made in the side of the mound to extract the charcoal. Despite a few unburnt brands the yield and quality was good filling almost 2 baskets.

To see if the kiln was reusable, I restacked it with timber cut from a fallen gum tree branch up the mountain. Due to the difficulty in reaching into the mound I stacked the wood in criss-crossed horizontal layers. The opening was sealed with mud and the mound lit as before. This time the mound burned quickly and I had to seal it early as the timber was burning at different rates, 3 hours after starting. Some large logs remained unburnt while charcoal that had already formed started to burn up being wasted as ash.

When I opened it the next day it had still produced an ok amount of charcoal but was disappointingly low compared to the first batch. This may partly be due to some of the wood being still green though it’s probably more likely to be due to how it was stacked. The lesson here is that when making charcoal the wood needs to be tightly stacked with few air spaces between. If not, the mound admits too much oxygen that quickly burns the timber.

Another thought I had was that wood may convert to charcoal better if laid vertically (or roughly so, like the cone in the first firing) so that the fire starts at the top of the wood and burns down. Stacking the wood in horizontal layers means that each layer has to set the one bellow alight leading to problems if the wood is green (use dry wood if stacking horizontally). By stacking wood vertically each piece is alight already and simply burns down towards the air entries. Stacking in this way also makes it easier to see fire in the air entries letting you know when to seal the mound.

For the reasons above I may make another charcoal kiln in future in the shape of a cylinder with air entries around the base and an open top. The kiln would be re-usable and easily stacked. A conical pile of wood would protrude above the walls of the kiln and be plastered in a temporary cover of mud. The kiln would be fired as with a normal mound and when finished the temporary cover of mud would be removed to extract the charcoal

Water Powered Hammer (Monjolo)

I built a water powered hammer called a “Monjolo”. I started by making a water spout from half a hollow log to direct water from the creek. This was set up in the creek and water flowed through it. The hammer was made from a fallen tree. I cut it to size by burning it at the points I wanted it cut (to save effort chopping). Next I carved a trough in one end to catch falling water. This was done first with a stone chisel that was then hafted to an L–shaped handle and used as an adze. This adze only took about an hour to make as I already had the chisel head and cordage made of bark fibre to bind it with.

To save further effort carving I used hot coals from the fire to char the wood in the trough. I put the coals in using “chopsticks” (unused arrow shafts) to transfer them from the pit. The coals were fanned or blown with a wooden blowpipe till the wood in the trough burned. Then the char was scraped out. The sides of the trough were sealed with clay to make sure the wooden sides did not burn away which would effectively decrease the volume of the trough. This was approximately 8 hours work over two days.

With the trough carved I made a hole in the middle of the log as a pivot point. Using the same char and scrape method I burnt a hole right through the log using hot coals and a blow pipe. Again clay was used to prevent wood burning where it was wanted. To burn through the approximately 25 cm diameter log it took about 4 hours and 30 minutes. Another hole was burnt in the end to fit the wooden hammer head and it took a similar amount of time.

A tripod lashed with loya cane was set up at the water spout. The axel of the hammer was tied to one leg, the hammer fitted onto the axel and the other end of the axel tied to another leg. The trough was positioned under the waterspout to collect water and the tripod adjusted so that the resting point of the hammer was horizontal (so water wouldn’t prematurely spill out of the trough).

The trough filled with water, outweighed the hammer head and tilted the hammer up into the air. The water then emptied out of the trough (now slanting downwards) and the hammer then slammed down onto an anvil stone returning to its original position. The cycle then repeated at the approximate rate of one strike every 10 seconds. The hammer crushes small soft types of stone like sandstone or ochre. I carved a bowl into the anvil stone so that it would collect the powder. I then crushed old pottery (useful as grog for new pots) and charcoal. Practically speaking, this hammer worked ok as a proof of concept but I might adjust it or make a new one with a larger trough and bigger hammer for heavy duty work.

This is the first machine I’ve built using primitive technology that produces work without human effort. Falling water replaces human calories to perform a repetitive task. A permanent set up usually has a shed protecting the hammer and materials from the weather while the trough end sits outside under the spout.  This type of hammer is used to pulverise grain into flour and I thought I might use one to mill dry cassava chips into flour when the garden matures. This device has also been used to crush clay for porcelain production. A stone head might make it useful as a stamp mill for crushing ores to powder. It might pulp fibres for paper even.

Termite clay kiln and pottery

I built this pottery kiln and some pottery from termite mound clay to test an alternative clay source to my usual one from the creek bank. I started by making a large grate from ordinary clay. It was just under 50 cm in diameter. Next, I took dry chunks of termite nest and put them into the pit in front of the tiled roof hut. The chunks were crushed and water was added to slake the clay. The clay was trodden on to mix it. Dead palm fronds were added to the clay to stop it from cracking as it dried and to add insulation to the kiln. The mixture was trodden on again and then taken from the pit. A trench was dug to form the firebox of the kiln and a wall of clay was made in the front of the trench. A hole was dug into the wall to allow air flow into the firebox.

The grate was placed on top of the firebox and the walls of the ware chamber were built around the grate. When the kiln walls were finished, grate bars made from termite clay were placed into the firebox. Grate bars are important for fireboxes as they lift the firewood off the ground allowing air to move up through the fuel bed for more efficient combustion. Burning wood as a heap on the ground allows cold air to flow up and over the coals, cooling the kiln and leaving the air unreacted with the fire wood. It still works but is much less efficient than using grate bars. The finished kiln was 50 cm tall (above grate height), 50 cm in diameter and with walls about 12.5 cm thick. The pit/firebox was about 25 cm deep and 25 cm wide with grate bars sitting half way between the ground and the circular kiln grate above.

Next, for the pottery clay, I selected a termite mound built on red clay soil. I took it to the kiln area and slaked it with water and mixed it in a small pit. I crushed up an old grate from a previous kiln and mixed it into the termite clay as grog. Grog prevents pottery from cracking as it dries and helps prevent breakage when firing. I then shaped the clay into a small urn. I also made some barrel roof tiles and a smaller pot from termite clay. I then stacked the kiln with the termite pottery (the urn, small pot and 5 barrel tiles) and some pottery made from normal clay (the housing for the forge blower and 2 barrel tiles).

To fire the pottery, I collected a large pile of dead wood and started a fire in the firebox. I heard some explosions in the kiln early on and knew something broke but continued anyway. To prevent explosions you should make sure all the pots are completely dry and slowly heat the kiln. Within an hour the kiln had heated up well and the pottery was glowing red hot. By the second hour the temperature went down illustrating an important point: if you over fill the firebox with wood the kiln will choke it and not burn efficiently. Realising this mistake I merely let the wood burn down a little so more air could get through. It’s important to watch the inside of the kiln and see how hot it’s glowing, try adding more or less wood and observe the effect on temperature. By 2 hours and 30 minutes the kiln was firing nicely again with all the pottery glowing low orange (about 845 c or 1550 f). I kept it at this low firing temperature for another 30 minutes. The whole firing process took about 3 hours from start to finish, a relatively short period of time for firing pottery.

When I took the pottery out, one tile had broken and the urn had spalled (a piece of the outer pot broke off) possibly due to still having moisture in it. The urn was still useable though and I use it to water the cassava patch. The forge blower was well fired and is now immune to water damage, no longer needing to be carefully protected from the rain. I put it in the barrel tile shed for storage. I put the broken tile and spalled piece from the urn in a special heap of broken pottery. When I make pottery in future I can crush up these broken pots and mix it into the new clay as grog to strengthen the new ceramic items. Finally, I stored the good tiles at the barrel tiled hut as replacements for broken tiles in that structure should there be any damage in future.

Termite clay is good material for making furnaces and an OK substitute for good pottery clay should it be difficult to find a better source. The termites have already processed the clay by the fact that their mouths are too small to include sticks and pebbles into their structures. As a result, the clay is very smooth and plastic. Too smooth for my liking, in fact, I’m used to working with coarser clay that has silt mixed into it naturally. I find that termite clay is either too runny when wet or cracks too easily when drier. It was difficult to form into complex shapes and it took me 2 attempts to make the urn. But for forming objects like tiles it’s OK, it can be pressed into shape and it will hold without difficulty. In future, I’d be likely to use termite clay for mass producing formed objects such as bricks, tiles, simple pots (formed over a mould) and possibly pipes, thereby conserving the dwindling clay supply from the creek bank which I’ll save for more intricate pottery. In summary, termite clay is able to be used to produce basic pottery if no other source can be found. If you have a termite nest you can make basic pottery from it.

Planting Cassava and Yams

In this video I build a garden to grow Cassava and yams, two staple food crops. Cassava is a shrub that develops large edible roots. Yams are a climbing vine that produce large, edible underground bulbs and smaller aerial bulbs on their vines.

I had 5 huts, but the wattle and daub hut (from the first video uploaded on this channel nearly 2 years ago) became dilapidated. I abandoned it in favour of the other huts I built and neglected the roof. This let water in destroying a wall. Also, the sweet potato patch behind it had a tree fall across it destroying the fence. So I demolished them both to make one large garden.

After removing the fence I set a fire under the fallen tree to burn it in half rather than spend the effort of cutting it with stone tools. After burning almost all the way through, it rained. So I came back later and cut through the rest of the log with stone tools. I eventually broke the tree in half. Using smaller logs as levers I moved the tree out of the garden clearing the space for the garden.

I then collected wood and built a simple fence that was woven loosely together with vine. The fence needs only to discourage large animals from entering to prevent them causing damage. Most times pigs and wallabies don’t know that food in growing in the garden and won’t try and enter if they see no reason to. Or at least that worked for the sweet potatoes so we’ll see if it works this time.

For the yam and cassava planting material I travelled far down stream to the site of my old stone hut that I built over 10 years ago. It had a corbelled dome roof that was damaged when a tree fell on it during a cyclone and it came down a few months later. The thick walls however have stayed standing for about a decade though.

Yams and cassava grew wild at this site which is one of the reasons I built the stone hut there. These plants are not native to Australia but grow wild here after having escaped from people’s gardens (similar to how wild pigs live here now after escaping from farms). The planting material for the yams are the bulbs that grow on the vines. The planting material for cassava are simply 25 cm long pieces of stem.

On returning to the garden, a scrub turkey was seen digging in the mounds. Protected by law, this bird has lost its fear of humans and in this case I’ve semi-domesticated it. Originally it was attracted to soil I dug up for the worms it exposed. I started leaving a pot out with small sweet potatoes in it for it to eat and now it investigates any pottery I leave for food. Now it my visits my projects and will only leave if bored or chased away. I suppose this is similar to how chickens were domesticated, in fact bush turkeys and chickens are related and will produce hybrid offspring.

Unfortunately, it has learned that the garden contains food. Originally, I was only going to plant yams but I saw the turkey digging them up and eating them. So, I planted cassava in the mounds so that the turkey would be discouraged by finding only wooden stems to peck at. I secretly planted the yams along the fence of the garden because the turkey only thinks the mounds contain yams. They can’t smell very well and only find food by sight and learned behaviour.

I planted the cassava in mounds 1 meter apart by pushing them flat into the soil. I planted the yams at intervals along the fence so they could use it as a trellis. 32 cassava stems and 12 yams were planted. Then a storm began and watered the garden. In less than a week the cassava had sprouted shoots and began to grow. The yams will take longer as I planted them deeper.

Cassava produces the most calories per time and space of any plant apart from sugar cane and sugar beet. But it requires much less fertiliser and effort. A hectare of cassava produces enough calories in 2 days to sustain a person for 1 year. It takes a year to come to harvest but will stay in the ground for a year without becoming woody. The tubers are high in starch and are what tapioca is made from.

This variety is called sweet cassava (actually not bitter cassava, it doesn’t taste sweet but starchy instead) and it needs to be boiled for 20 minutes to get rid of some cyanide it contains. The bitter variety contains such high levels that it kills if eaten raw and requires more extensive treatment to eat. There isn’t much nutrition in cassava other than the large amount calories it contains so other food would be required to provide protein and nutrients.

After I harvest the cassava I planted I’ll try fermenting it (which adds nutrition), drying it and pounding it into flour to make flat bread. Cassava flour has the same energy content as wheat flour, stores well and tastes somewhat similar. Or I could just cook it and eat it straight from the garden. I’ll use the yams like potatoes when they’re ready.

source:https://research.googleblog.com/2017/12/improving-end-to-end-models-for-speech.html