axonml 0.6.0

A complete ML/AI framework in pure Rust - PyTorch-equivalent functionality
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283

//! AQUILO — Electrical Systems Diagnostic Model (~608K params)
//!
//! # File
//! `crates/axonml/src/hvac/aquilo.rs`
//!
//! # Author
//! Andrew Jewell Sr - AutomataNexus
//!
//! # Updated
//! March 8, 2026
//!
//! # Disclaimer
//! Use at own risk. This software is provided "as is", without warranty of any
//! kind, express or implied. The author and AutomataNexus shall not be held
//! liable for any damages arising from the use of this software.

use std::collections::HashMap;

use axonml_autograd::Variable;
use axonml_nn::{BatchNorm1d, Dropout, FFT1d, GELU, Linear, Module, Parameter, ReLU, Sequential};

// =============================================================================
// Aquilo Model
// =============================================================================

/// Electrical systems diagnostic model.
///
/// Architecture:
/// - 3 parallel analyzers (voltage, current, power quality)
/// - FFT spectral feature extraction on voltage/current
/// - Fusion network with multi-head output
///
/// Input: (batch, 168) — flattened 64 timesteps × 7 features, split 3×56
/// Outputs: fault(13), severity(5), phase_health(3), power_quality(1)
pub struct Aquilo {
    voltage_analyzer: Sequential,
    current_analyzer: Sequential,
    power_quality: Sequential,
    fft: FFT1d,
    main_net: Sequential,
    fault_head: Linear,
    severity_head: Linear,
    phase_health_head: Linear,
    pq_head: Linear,
    training: bool,
}

impl Default for Aquilo {
    fn default() -> Self {
        Self::new()
    }
}

impl Aquilo {
    /// Creates a new Aquilo model.
    pub fn new() -> Self {
        let voltage_analyzer = Sequential::new()
            .add(Linear::new(56, 64))
            .add(BatchNorm1d::new(64))
            .add(ReLU)
            .add(Dropout::new(0.2))
            .add(Linear::new(64, 64));

        let current_analyzer = Sequential::new()
            .add(Linear::new(56, 64))
            .add(BatchNorm1d::new(64))
            .add(ReLU)
            .add(Dropout::new(0.2))
            .add(Linear::new(64, 64));

        let power_quality = Sequential::new()
            .add(Linear::new(56, 32))
            .add(ReLU)
            .add(Linear::new(32, 32));

        // FFT on 56-length segments → 29 bins each, 2 segments = 58 features
        let fft = FFT1d::new(56);

        // Concat: raw(168) + voltage(64) + current(64) + pq(32) + fft(58) = 386
        let main_net = Sequential::new()
            .add(Linear::new(386, 512))
            .add(BatchNorm1d::new(512))
            .add(GELU)
            .add(Dropout::new(0.3))
            .add(Linear::new(512, 256))
            .add(BatchNorm1d::new(256))
            .add(ReLU)
            .add(Dropout::new(0.2));

        let fault_head = Linear::new(256, 13);
        let severity_head = Linear::new(256, 5);
        let phase_health_head = Linear::new(256, 3);
        let pq_head = Linear::new(256, 1);

        Self {
            voltage_analyzer,
            current_analyzer,
            power_quality,
            fft,
            main_net,
            fault_head,
            severity_head,
            phase_health_head,
            pq_head,
            training: true,
        }
    }

    /// Forward pass returning all output heads.
    ///
    /// Returns (fault_logits, severity_logits, phase_health, power_quality, embedding)
    pub fn forward_all(
        &self,
        input: &Variable,
    ) -> (Variable, Variable, Variable, Variable, Variable) {
        let shape = input.shape();
        let batch = shape[0];

        // Split input into 3 segments of 56 using narrow (GPU-compatible)
        let volt_var = input.narrow(1, 0, 56); // (batch, 56)
        let curr_var = input.narrow(1, 56, 56); // (batch, 56)
        let pq_var = input.narrow(1, 112, 56); // (batch, 56)

        // Analyzer branches
        let volt_out = self.voltage_analyzer.forward(&volt_var); // (batch, 64)
        let curr_out = self.current_analyzer.forward(&curr_var); // (batch, 64)
        let pq_out = self.power_quality.forward(&pq_var); // (batch, 32)

        // FFT spectral features on voltage and current
        let volt_fft = self.fft.forward(&volt_var); // (batch, 29)
        let curr_fft = self.fft.forward(&curr_var); // (batch, 29)

        // Concatenate everything: raw(168) + volt(64) + curr(64) + pq(32) + fft(58) = 386
        let all_features = concat_variables(
            &[input, &volt_out, &curr_out, &pq_out, &volt_fft, &curr_fft],
            batch,
        );

        // Main network
        let embedding = self.main_net.forward(&all_features); // (batch, 256)

        // Output heads
        let fault = self.fault_head.forward(&embedding);
        let severity = self.severity_head.forward(&embedding);
        let phase_health = self.phase_health_head.forward(&embedding);
        let power_quality = self.pq_head.forward(&embedding);

        (fault, severity, phase_health, power_quality, embedding)
    }

    /// Returns the embedding dimension for downstream aggregators.
    pub fn embedding_dim() -> usize {
        256
    }

    /// Returns total output dimension (13 + 5 + 3 + 1 = 22).
    pub fn output_dim() -> usize {
        22
    }
}

impl Module for Aquilo {
    fn forward(&self, input: &Variable) -> Variable {
        let (fault, _, _, _, _) = self.forward_all(input);
        fault
    }

    fn parameters(&self) -> Vec<Parameter> {
        let mut params = Vec::new();
        params.extend(self.voltage_analyzer.parameters());
        params.extend(self.current_analyzer.parameters());
        params.extend(self.power_quality.parameters());
        params.extend(self.main_net.parameters());
        params.extend(self.fault_head.parameters());
        params.extend(self.severity_head.parameters());
        params.extend(self.phase_health_head.parameters());
        params.extend(self.pq_head.parameters());
        params
    }

    fn named_parameters(&self) -> HashMap<String, Parameter> {
        let mut params = HashMap::new();
        for (n, p) in self.voltage_analyzer.named_parameters() {
            params.insert(format!("voltage_analyzer.{n}"), p);
        }
        for (n, p) in self.current_analyzer.named_parameters() {
            params.insert(format!("current_analyzer.{n}"), p);
        }
        for (n, p) in self.power_quality.named_parameters() {
            params.insert(format!("power_quality.{n}"), p);
        }
        for (n, p) in self.main_net.named_parameters() {
            params.insert(format!("main_net.{n}"), p);
        }
        for (n, p) in self.fault_head.named_parameters() {
            params.insert(format!("fault_head.{n}"), p);
        }
        for (n, p) in self.severity_head.named_parameters() {
            params.insert(format!("severity_head.{n}"), p);
        }
        for (n, p) in self.phase_health_head.named_parameters() {
            params.insert(format!("phase_health_head.{n}"), p);
        }
        for (n, p) in self.pq_head.named_parameters() {
            params.insert(format!("pq_head.{n}"), p);
        }
        params
    }

    fn set_training(&mut self, training: bool) {
        self.training = training;
        self.voltage_analyzer.set_training(training);
        self.current_analyzer.set_training(training);
        self.power_quality.set_training(training);
        self.main_net.set_training(training);
    }

    fn is_training(&self) -> bool {
        self.training
    }

    fn name(&self) -> &'static str {
        "Aquilo"
    }
}

// =============================================================================
// Utility
// =============================================================================

/// Concatenate multiple Variables along the last dimension.
pub(crate) fn concat_variables(vars: &[&Variable], _batch: usize) -> Variable {
    Variable::cat(vars, vars[0].shape().len() - 1)
}

// =============================================================================
// Tests
// =============================================================================

#[cfg(test)]
mod tests {
    use super::*;
    use axonml_tensor::Tensor;

    #[test]
    fn test_aquilo_output_shapes() {
        let model = Aquilo::new();
        let input = Variable::new(
            Tensor::from_vec(vec![1.0; 2 * 168], &[2, 168]).unwrap(),
            false,
        );
        let (fault, severity, phase, pq, emb) = model.forward_all(&input);

        assert_eq!(fault.shape(), vec![2, 13]);
        assert_eq!(severity.shape(), vec![2, 5]);
        assert_eq!(phase.shape(), vec![2, 3]);
        assert_eq!(pq.shape(), vec![2, 1]);
        assert_eq!(emb.shape(), vec![2, 256]);
    }

    #[test]
    fn test_aquilo_parameter_count() {
        let model = Aquilo::new();
        let total: usize = model.parameters().iter().map(|p| p.numel()).sum();
        // Architecture yields ~355K params
        assert!(
            total > 250_000 && total < 500_000,
            "Aquilo has {} params, expected ~355K",
            total
        );
    }

    #[test]
    fn test_aquilo_forward_module_trait() {
        let model = Aquilo::new();
        let input = Variable::new(
            Tensor::from_vec(vec![1.0; 4 * 168], &[4, 168]).unwrap(),
            false,
        );
        let output = model.forward(&input);
        assert_eq!(output.shape(), vec![4, 13]);
    }
}